2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "ref-cache.h"
35 #define PENDING_EXTENT_INSERT 0
36 #define PENDING_EXTENT_DELETE 1
37 #define PENDING_BACKREF_UPDATE 2
39 struct pending_extent_op {
48 struct list_head list;
52 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *root, u64 parent,
54 u64 root_objectid, u64 ref_generation,
55 u64 owner, struct btrfs_key *ins,
57 static int update_reserved_extents(struct btrfs_root *root,
58 u64 bytenr, u64 num, int reserve);
59 static int pin_down_bytes(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int is_data);
62 static int update_block_group(struct btrfs_trans_handle *trans,
63 struct btrfs_root *root,
64 u64 bytenr, u64 num_bytes, int alloc,
66 static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
67 struct btrfs_root *root,
68 u64 bytenr, u64 num_bytes, u64 parent,
69 u64 root_objectid, u64 ref_generation,
70 u64 owner_objectid, int pin,
73 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
74 struct btrfs_root *extent_root, u64 alloc_bytes,
75 u64 flags, int force);
77 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
79 return (cache->flags & bits) == bits;
83 * this adds the block group to the fs_info rb tree for the block group
86 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
87 struct btrfs_block_group_cache *block_group)
90 struct rb_node *parent = NULL;
91 struct btrfs_block_group_cache *cache;
93 spin_lock(&info->block_group_cache_lock);
94 p = &info->block_group_cache_tree.rb_node;
98 cache = rb_entry(parent, struct btrfs_block_group_cache,
100 if (block_group->key.objectid < cache->key.objectid) {
102 } else if (block_group->key.objectid > cache->key.objectid) {
105 spin_unlock(&info->block_group_cache_lock);
110 rb_link_node(&block_group->cache_node, parent, p);
111 rb_insert_color(&block_group->cache_node,
112 &info->block_group_cache_tree);
113 spin_unlock(&info->block_group_cache_lock);
119 * This will return the block group at or after bytenr if contains is 0, else
120 * it will return the block group that contains the bytenr
122 static struct btrfs_block_group_cache *
123 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
126 struct btrfs_block_group_cache *cache, *ret = NULL;
130 spin_lock(&info->block_group_cache_lock);
131 n = info->block_group_cache_tree.rb_node;
134 cache = rb_entry(n, struct btrfs_block_group_cache,
136 end = cache->key.objectid + cache->key.offset - 1;
137 start = cache->key.objectid;
139 if (bytenr < start) {
140 if (!contains && (!ret || start < ret->key.objectid))
143 } else if (bytenr > start) {
144 if (contains && bytenr <= end) {
155 atomic_inc(&ret->count);
156 spin_unlock(&info->block_group_cache_lock);
162 * this is only called by cache_block_group, since we could have freed extents
163 * we need to check the pinned_extents for any extents that can't be used yet
164 * since their free space will be released as soon as the transaction commits.
166 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
167 struct btrfs_fs_info *info, u64 start, u64 end)
169 u64 extent_start, extent_end, size;
172 mutex_lock(&info->pinned_mutex);
173 while (start < end) {
174 ret = find_first_extent_bit(&info->pinned_extents, start,
175 &extent_start, &extent_end,
180 if (extent_start == start) {
181 start = extent_end + 1;
182 } else if (extent_start > start && extent_start < end) {
183 size = extent_start - start;
184 ret = btrfs_add_free_space(block_group, start,
187 start = extent_end + 1;
195 ret = btrfs_add_free_space(block_group, start, size);
198 mutex_unlock(&info->pinned_mutex);
203 static int remove_sb_from_cache(struct btrfs_root *root,
204 struct btrfs_block_group_cache *cache)
211 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
212 bytenr = btrfs_sb_offset(i);
213 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
214 cache->key.objectid, bytenr, 0,
215 &logical, &nr, &stripe_len);
218 btrfs_remove_free_space(cache, logical[nr],
226 static int cache_block_group(struct btrfs_root *root,
227 struct btrfs_block_group_cache *block_group)
229 struct btrfs_path *path;
231 struct btrfs_key key;
232 struct extent_buffer *leaf;
239 root = root->fs_info->extent_root;
241 if (block_group->cached)
244 path = btrfs_alloc_path();
250 * we get into deadlocks with paths held by callers of this function.
251 * since the alloc_mutex is protecting things right now, just
252 * skip the locking here
254 path->skip_locking = 1;
255 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
258 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
259 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
264 leaf = path->nodes[0];
265 slot = path->slots[0];
266 if (slot >= btrfs_header_nritems(leaf)) {
267 ret = btrfs_next_leaf(root, path);
275 btrfs_item_key_to_cpu(leaf, &key, slot);
276 if (key.objectid < block_group->key.objectid)
279 if (key.objectid >= block_group->key.objectid +
280 block_group->key.offset)
283 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
284 add_new_free_space(block_group, root->fs_info, last,
287 last = key.objectid + key.offset;
293 add_new_free_space(block_group, root->fs_info, last,
294 block_group->key.objectid +
295 block_group->key.offset);
297 remove_sb_from_cache(root, block_group);
298 block_group->cached = 1;
301 btrfs_free_path(path);
306 * return the block group that starts at or after bytenr
308 static struct btrfs_block_group_cache *
309 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
311 struct btrfs_block_group_cache *cache;
313 cache = block_group_cache_tree_search(info, bytenr, 0);
319 * return the block group that contains teh given bytenr
321 struct btrfs_block_group_cache *btrfs_lookup_block_group(
322 struct btrfs_fs_info *info,
325 struct btrfs_block_group_cache *cache;
327 cache = block_group_cache_tree_search(info, bytenr, 1);
332 static inline void put_block_group(struct btrfs_block_group_cache *cache)
334 if (atomic_dec_and_test(&cache->count))
338 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
341 struct list_head *head = &info->space_info;
342 struct btrfs_space_info *found;
345 list_for_each_entry_rcu(found, head, list) {
346 if (found->flags == flags) {
356 * after adding space to the filesystem, we need to clear the full flags
357 * on all the space infos.
359 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
361 struct list_head *head = &info->space_info;
362 struct btrfs_space_info *found;
365 list_for_each_entry_rcu(found, head, list)
370 static u64 div_factor(u64 num, int factor)
379 u64 btrfs_find_block_group(struct btrfs_root *root,
380 u64 search_start, u64 search_hint, int owner)
382 struct btrfs_block_group_cache *cache;
384 u64 last = max(search_hint, search_start);
391 cache = btrfs_lookup_first_block_group(root->fs_info, last);
395 spin_lock(&cache->lock);
396 last = cache->key.objectid + cache->key.offset;
397 used = btrfs_block_group_used(&cache->item);
399 if ((full_search || !cache->ro) &&
400 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
401 if (used + cache->pinned + cache->reserved <
402 div_factor(cache->key.offset, factor)) {
403 group_start = cache->key.objectid;
404 spin_unlock(&cache->lock);
405 put_block_group(cache);
409 spin_unlock(&cache->lock);
410 put_block_group(cache);
418 if (!full_search && factor < 10) {
428 /* simple helper to search for an existing extent at a given offset */
429 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
432 struct btrfs_key key;
433 struct btrfs_path *path;
435 path = btrfs_alloc_path();
437 key.objectid = start;
439 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
440 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
442 btrfs_free_path(path);
447 * Back reference rules. Back refs have three main goals:
449 * 1) differentiate between all holders of references to an extent so that
450 * when a reference is dropped we can make sure it was a valid reference
451 * before freeing the extent.
453 * 2) Provide enough information to quickly find the holders of an extent
454 * if we notice a given block is corrupted or bad.
456 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
457 * maintenance. This is actually the same as #2, but with a slightly
458 * different use case.
460 * File extents can be referenced by:
462 * - multiple snapshots, subvolumes, or different generations in one subvol
463 * - different files inside a single subvolume
464 * - different offsets inside a file (bookend extents in file.c)
466 * The extent ref structure has fields for:
468 * - Objectid of the subvolume root
469 * - Generation number of the tree holding the reference
470 * - objectid of the file holding the reference
471 * - number of references holding by parent node (alway 1 for tree blocks)
473 * Btree leaf may hold multiple references to a file extent. In most cases,
474 * these references are from same file and the corresponding offsets inside
475 * the file are close together.
477 * When a file extent is allocated the fields are filled in:
478 * (root_key.objectid, trans->transid, inode objectid, 1)
480 * When a leaf is cow'd new references are added for every file extent found
481 * in the leaf. It looks similar to the create case, but trans->transid will
482 * be different when the block is cow'd.
484 * (root_key.objectid, trans->transid, inode objectid,
485 * number of references in the leaf)
487 * When a file extent is removed either during snapshot deletion or
488 * file truncation, we find the corresponding back reference and check
489 * the following fields:
491 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
494 * Btree extents can be referenced by:
496 * - Different subvolumes
497 * - Different generations of the same subvolume
499 * When a tree block is created, back references are inserted:
501 * (root->root_key.objectid, trans->transid, level, 1)
503 * When a tree block is cow'd, new back references are added for all the
504 * blocks it points to. If the tree block isn't in reference counted root,
505 * the old back references are removed. These new back references are of
506 * the form (trans->transid will have increased since creation):
508 * (root->root_key.objectid, trans->transid, level, 1)
510 * When a backref is in deleting, the following fields are checked:
512 * if backref was for a tree root:
513 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
515 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
517 * Back Reference Key composing:
519 * The key objectid corresponds to the first byte in the extent, the key
520 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
521 * byte of parent extent. If a extent is tree root, the key offset is set
522 * to the key objectid.
525 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
526 struct btrfs_root *root,
527 struct btrfs_path *path,
528 u64 bytenr, u64 parent,
529 u64 ref_root, u64 ref_generation,
530 u64 owner_objectid, int del)
532 struct btrfs_key key;
533 struct btrfs_extent_ref *ref;
534 struct extent_buffer *leaf;
538 key.objectid = bytenr;
539 key.type = BTRFS_EXTENT_REF_KEY;
542 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
550 leaf = path->nodes[0];
551 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
552 ref_objectid = btrfs_ref_objectid(leaf, ref);
553 if (btrfs_ref_root(leaf, ref) != ref_root ||
554 btrfs_ref_generation(leaf, ref) != ref_generation ||
555 (ref_objectid != owner_objectid &&
556 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
566 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
567 struct btrfs_root *root,
568 struct btrfs_path *path,
569 u64 bytenr, u64 parent,
570 u64 ref_root, u64 ref_generation,
574 struct btrfs_key key;
575 struct extent_buffer *leaf;
576 struct btrfs_extent_ref *ref;
580 key.objectid = bytenr;
581 key.type = BTRFS_EXTENT_REF_KEY;
584 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
586 leaf = path->nodes[0];
587 ref = btrfs_item_ptr(leaf, path->slots[0],
588 struct btrfs_extent_ref);
589 btrfs_set_ref_root(leaf, ref, ref_root);
590 btrfs_set_ref_generation(leaf, ref, ref_generation);
591 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
592 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
593 } else if (ret == -EEXIST) {
596 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
597 leaf = path->nodes[0];
598 ref = btrfs_item_ptr(leaf, path->slots[0],
599 struct btrfs_extent_ref);
600 if (btrfs_ref_root(leaf, ref) != ref_root ||
601 btrfs_ref_generation(leaf, ref) != ref_generation) {
607 num_refs = btrfs_ref_num_refs(leaf, ref);
608 BUG_ON(num_refs == 0);
609 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
611 existing_owner = btrfs_ref_objectid(leaf, ref);
612 if (existing_owner != owner_objectid &&
613 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
614 btrfs_set_ref_objectid(leaf, ref,
615 BTRFS_MULTIPLE_OBJECTIDS);
621 btrfs_mark_buffer_dirty(path->nodes[0]);
623 btrfs_release_path(root, path);
627 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
628 struct btrfs_root *root,
629 struct btrfs_path *path,
632 struct extent_buffer *leaf;
633 struct btrfs_extent_ref *ref;
637 leaf = path->nodes[0];
638 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
639 num_refs = btrfs_ref_num_refs(leaf, ref);
640 BUG_ON(num_refs < refs_to_drop);
641 num_refs -= refs_to_drop;
643 ret = btrfs_del_item(trans, root, path);
645 btrfs_set_ref_num_refs(leaf, ref, num_refs);
646 btrfs_mark_buffer_dirty(leaf);
648 btrfs_release_path(root, path);
652 #ifdef BIO_RW_DISCARD
653 static void btrfs_issue_discard(struct block_device *bdev,
656 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
660 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
663 #ifdef BIO_RW_DISCARD
665 u64 map_length = num_bytes;
666 struct btrfs_multi_bio *multi = NULL;
668 /* Tell the block device(s) that the sectors can be discarded */
669 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
670 bytenr, &map_length, &multi, 0);
672 struct btrfs_bio_stripe *stripe = multi->stripes;
675 if (map_length > num_bytes)
676 map_length = num_bytes;
678 for (i = 0; i < multi->num_stripes; i++, stripe++) {
679 btrfs_issue_discard(stripe->dev->bdev,
692 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
693 struct btrfs_root *root, u64 bytenr,
695 u64 orig_parent, u64 parent,
696 u64 orig_root, u64 ref_root,
697 u64 orig_generation, u64 ref_generation,
701 int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
703 ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
704 orig_parent, parent, orig_root,
705 ref_root, orig_generation,
706 ref_generation, owner_objectid, pin);
711 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
712 struct btrfs_root *root, u64 bytenr,
713 u64 num_bytes, u64 orig_parent, u64 parent,
714 u64 ref_root, u64 ref_generation,
718 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
719 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
722 ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
723 orig_parent, parent, ref_root,
724 ref_root, ref_generation,
725 ref_generation, owner_objectid);
728 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
729 struct btrfs_root *root, u64 bytenr,
731 u64 orig_parent, u64 parent,
732 u64 orig_root, u64 ref_root,
733 u64 orig_generation, u64 ref_generation,
738 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
739 ref_generation, owner_objectid,
740 BTRFS_ADD_DELAYED_REF, 0);
745 static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
746 struct btrfs_root *root, u64 bytenr,
747 u64 num_bytes, u64 parent, u64 ref_root,
748 u64 ref_generation, u64 owner_objectid,
751 struct btrfs_path *path;
753 struct btrfs_key key;
754 struct extent_buffer *l;
755 struct btrfs_extent_item *item;
758 path = btrfs_alloc_path();
763 key.objectid = bytenr;
764 key.type = BTRFS_EXTENT_ITEM_KEY;
765 key.offset = num_bytes;
767 /* first find the extent item and update its reference count */
768 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
775 btrfs_free_path(path);
780 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
781 if (key.objectid != bytenr) {
782 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
783 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
784 (unsigned long long)bytenr,
785 (unsigned long long)key.objectid);
788 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
790 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
792 refs = btrfs_extent_refs(l, item);
793 btrfs_set_extent_refs(l, item, refs + refs_to_add);
794 btrfs_mark_buffer_dirty(path->nodes[0]);
796 btrfs_release_path(root->fs_info->extent_root, path);
799 /* now insert the actual backref */
800 ret = insert_extent_backref(trans, root->fs_info->extent_root,
801 path, bytenr, parent,
802 ref_root, ref_generation,
803 owner_objectid, refs_to_add);
805 btrfs_free_path(path);
809 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root,
811 u64 bytenr, u64 num_bytes, u64 parent,
812 u64 ref_root, u64 ref_generation,
816 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
817 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
820 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
821 0, ref_root, 0, ref_generation,
826 static int drop_delayed_ref(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_delayed_ref_node *node)
831 struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
833 BUG_ON(node->ref_mod == 0);
834 ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
835 node->parent, ref->root, ref->generation,
836 ref->owner_objectid, ref->pin, node->ref_mod);
841 /* helper function to actually process a single delayed ref entry */
842 static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
843 struct btrfs_root *root,
844 struct btrfs_delayed_ref_node *node,
848 struct btrfs_delayed_ref *ref;
850 if (node->parent == (u64)-1) {
851 struct btrfs_delayed_ref_head *head;
853 * we've hit the end of the chain and we were supposed
854 * to insert this extent into the tree. But, it got
855 * deleted before we ever needed to insert it, so all
856 * we have to do is clean up the accounting
858 if (insert_reserved) {
859 update_reserved_extents(root, node->bytenr,
862 head = btrfs_delayed_node_to_head(node);
863 mutex_unlock(&head->mutex);
867 ref = btrfs_delayed_node_to_ref(node);
868 if (ref->action == BTRFS_ADD_DELAYED_REF) {
869 if (insert_reserved) {
870 struct btrfs_key ins;
872 ins.objectid = node->bytenr;
873 ins.offset = node->num_bytes;
874 ins.type = BTRFS_EXTENT_ITEM_KEY;
876 /* record the full extent allocation */
877 ret = __btrfs_alloc_reserved_extent(trans, root,
878 node->parent, ref->root,
879 ref->generation, ref->owner_objectid,
880 &ins, node->ref_mod);
881 update_reserved_extents(root, node->bytenr,
884 /* just add one backref */
885 ret = add_extent_ref(trans, root, node->bytenr,
887 node->parent, ref->root, ref->generation,
888 ref->owner_objectid, node->ref_mod);
891 } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
892 WARN_ON(insert_reserved);
893 ret = drop_delayed_ref(trans, root, node);
898 static noinline struct btrfs_delayed_ref_node *
899 select_delayed_ref(struct btrfs_delayed_ref_head *head)
901 struct rb_node *node;
902 struct btrfs_delayed_ref_node *ref;
903 int action = BTRFS_ADD_DELAYED_REF;
906 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
907 * this prevents ref count from going down to zero when
908 * there still are pending delayed ref.
910 node = rb_prev(&head->node.rb_node);
914 ref = rb_entry(node, struct btrfs_delayed_ref_node,
916 if (ref->bytenr != head->node.bytenr)
918 if (btrfs_delayed_node_to_ref(ref)->action == action)
920 node = rb_prev(node);
922 if (action == BTRFS_ADD_DELAYED_REF) {
923 action = BTRFS_DROP_DELAYED_REF;
930 * this starts processing the delayed reference count updates and
931 * extent insertions we have queued up so far. count can be
932 * 0, which means to process everything in the tree at the start
933 * of the run (but not newly added entries), or it can be some target
934 * number you'd like to process.
936 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
937 struct btrfs_root *root, unsigned long count)
939 struct rb_node *node;
940 struct btrfs_delayed_ref_root *delayed_refs;
941 struct btrfs_delayed_ref_node *ref;
942 struct btrfs_delayed_ref_head *locked_ref = NULL;
944 int must_insert_reserved = 0;
945 int run_all = count == (unsigned long)-1;
947 if (root == root->fs_info->extent_root)
948 root = root->fs_info->tree_root;
950 delayed_refs = &trans->transaction->delayed_refs;
952 spin_lock(&delayed_refs->lock);
954 count = delayed_refs->num_entries;
958 * no locked ref, go find something we can
959 * process in the rbtree. We start at
960 * the beginning of the tree, there may be less
961 * lock contention if we do something smarter here.
963 node = rb_first(&delayed_refs->root);
965 spin_unlock(&delayed_refs->lock);
969 ref = rb_entry(node, struct btrfs_delayed_ref_node,
971 ret = btrfs_lock_delayed_ref(trans, ref, &locked_ref);
973 spin_unlock(&delayed_refs->lock);
979 * record the must insert reserved flag before we
980 * drop the spin lock.
982 must_insert_reserved = locked_ref->must_insert_reserved;
983 locked_ref->must_insert_reserved = 0;
986 * locked_ref is the head node, so we have to go one
987 * node back for any delayed ref updates
990 ref = select_delayed_ref(locked_ref);
992 /* All delayed refs have been processed, Go ahead
993 * and send the head node to run_one_delayed_ref,
994 * so that any accounting fixes can happen
996 ref = &locked_ref->node;
1001 rb_erase(&ref->rb_node, &delayed_refs->root);
1002 delayed_refs->num_entries--;
1003 spin_unlock(&delayed_refs->lock);
1005 ret = run_one_delayed_ref(trans, root, ref,
1006 must_insert_reserved);
1008 btrfs_put_delayed_ref(ref);
1010 /* once we lock the head ref, we have to process all the
1011 * entries for it. So, we might end up doing more entries
1012 * that count was asking us to do.
1018 * we set locked_ref to null above if we're all done
1021 if (!locked_ref && count == 0)
1025 spin_lock(&delayed_refs->lock);
1028 spin_lock(&delayed_refs->lock);
1029 node = rb_first(&delayed_refs->root);
1031 spin_unlock(&delayed_refs->lock);
1036 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1038 if (btrfs_delayed_ref_is_head(ref)) {
1039 struct btrfs_delayed_ref_head *head;
1041 head = btrfs_delayed_node_to_head(ref);
1042 atomic_inc(&ref->refs);
1044 spin_unlock(&delayed_refs->lock);
1045 mutex_lock(&head->mutex);
1046 mutex_unlock(&head->mutex);
1048 btrfs_put_delayed_ref(ref);
1052 node = rb_next(node);
1054 spin_unlock(&delayed_refs->lock);
1055 count = (unsigned long)-1;
1056 schedule_timeout(1);
1063 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1064 struct btrfs_root *root, u64 objectid, u64 bytenr)
1066 struct btrfs_root *extent_root = root->fs_info->extent_root;
1067 struct btrfs_path *path;
1068 struct extent_buffer *leaf;
1069 struct btrfs_extent_ref *ref_item;
1070 struct btrfs_key key;
1071 struct btrfs_key found_key;
1077 key.objectid = bytenr;
1078 key.offset = (u64)-1;
1079 key.type = BTRFS_EXTENT_ITEM_KEY;
1081 path = btrfs_alloc_path();
1082 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1088 if (path->slots[0] == 0)
1092 leaf = path->nodes[0];
1093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1095 if (found_key.objectid != bytenr ||
1096 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1099 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1101 leaf = path->nodes[0];
1102 nritems = btrfs_header_nritems(leaf);
1103 if (path->slots[0] >= nritems) {
1104 ret = btrfs_next_leaf(extent_root, path);
1111 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1112 if (found_key.objectid != bytenr)
1115 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1120 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1121 struct btrfs_extent_ref);
1122 ref_root = btrfs_ref_root(leaf, ref_item);
1123 if ((ref_root != root->root_key.objectid &&
1124 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1125 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1129 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1138 btrfs_free_path(path);
1142 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1143 struct extent_buffer *buf, u32 nr_extents)
1145 struct btrfs_key key;
1146 struct btrfs_file_extent_item *fi;
1154 if (!root->ref_cows)
1157 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1159 root_gen = root->root_key.offset;
1162 root_gen = trans->transid - 1;
1165 level = btrfs_header_level(buf);
1166 nritems = btrfs_header_nritems(buf);
1169 struct btrfs_leaf_ref *ref;
1170 struct btrfs_extent_info *info;
1172 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1178 ref->root_gen = root_gen;
1179 ref->bytenr = buf->start;
1180 ref->owner = btrfs_header_owner(buf);
1181 ref->generation = btrfs_header_generation(buf);
1182 ref->nritems = nr_extents;
1183 info = ref->extents;
1185 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1187 btrfs_item_key_to_cpu(buf, &key, i);
1188 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1190 fi = btrfs_item_ptr(buf, i,
1191 struct btrfs_file_extent_item);
1192 if (btrfs_file_extent_type(buf, fi) ==
1193 BTRFS_FILE_EXTENT_INLINE)
1195 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1196 if (disk_bytenr == 0)
1199 info->bytenr = disk_bytenr;
1201 btrfs_file_extent_disk_num_bytes(buf, fi);
1202 info->objectid = key.objectid;
1203 info->offset = key.offset;
1207 ret = btrfs_add_leaf_ref(root, ref, shared);
1208 if (ret == -EEXIST && shared) {
1209 struct btrfs_leaf_ref *old;
1210 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1212 btrfs_remove_leaf_ref(root, old);
1213 btrfs_free_leaf_ref(root, old);
1214 ret = btrfs_add_leaf_ref(root, ref, shared);
1217 btrfs_free_leaf_ref(root, ref);
1223 /* when a block goes through cow, we update the reference counts of
1224 * everything that block points to. The internal pointers of the block
1225 * can be in just about any order, and it is likely to have clusters of
1226 * things that are close together and clusters of things that are not.
1228 * To help reduce the seeks that come with updating all of these reference
1229 * counts, sort them by byte number before actual updates are done.
1231 * struct refsort is used to match byte number to slot in the btree block.
1232 * we sort based on the byte number and then use the slot to actually
1235 * struct refsort is smaller than strcut btrfs_item and smaller than
1236 * struct btrfs_key_ptr. Since we're currently limited to the page size
1237 * for a btree block, there's no way for a kmalloc of refsorts for a
1238 * single node to be bigger than a page.
1246 * for passing into sort()
1248 static int refsort_cmp(const void *a_void, const void *b_void)
1250 const struct refsort *a = a_void;
1251 const struct refsort *b = b_void;
1253 if (a->bytenr < b->bytenr)
1255 if (a->bytenr > b->bytenr)
1261 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct extent_buffer *orig_buf,
1264 struct extent_buffer *buf, u32 *nr_extents)
1270 u64 orig_generation;
1271 struct refsort *sorted;
1273 u32 nr_file_extents = 0;
1274 struct btrfs_key key;
1275 struct btrfs_file_extent_item *fi;
1282 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1283 u64, u64, u64, u64, u64, u64, u64, u64, u64);
1285 ref_root = btrfs_header_owner(buf);
1286 ref_generation = btrfs_header_generation(buf);
1287 orig_root = btrfs_header_owner(orig_buf);
1288 orig_generation = btrfs_header_generation(orig_buf);
1290 nritems = btrfs_header_nritems(buf);
1291 level = btrfs_header_level(buf);
1293 sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1296 if (root->ref_cows) {
1297 process_func = __btrfs_inc_extent_ref;
1300 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1303 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1305 process_func = __btrfs_update_extent_ref;
1309 * we make two passes through the items. In the first pass we
1310 * only record the byte number and slot. Then we sort based on
1311 * byte number and do the actual work based on the sorted results
1313 for (i = 0; i < nritems; i++) {
1316 btrfs_item_key_to_cpu(buf, &key, i);
1317 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1319 fi = btrfs_item_ptr(buf, i,
1320 struct btrfs_file_extent_item);
1321 if (btrfs_file_extent_type(buf, fi) ==
1322 BTRFS_FILE_EXTENT_INLINE)
1324 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1329 sorted[refi].bytenr = bytenr;
1330 sorted[refi].slot = i;
1333 bytenr = btrfs_node_blockptr(buf, i);
1334 sorted[refi].bytenr = bytenr;
1335 sorted[refi].slot = i;
1340 * if refi == 0, we didn't actually put anything into the sorted
1341 * array and we're done
1346 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1348 for (i = 0; i < refi; i++) {
1350 slot = sorted[i].slot;
1351 bytenr = sorted[i].bytenr;
1354 btrfs_item_key_to_cpu(buf, &key, slot);
1355 fi = btrfs_item_ptr(buf, slot,
1356 struct btrfs_file_extent_item);
1358 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1362 ret = process_func(trans, root, bytenr,
1363 btrfs_file_extent_disk_num_bytes(buf, fi),
1364 orig_buf->start, buf->start,
1365 orig_root, ref_root,
1366 orig_generation, ref_generation,
1375 ret = process_func(trans, root, bytenr, buf->len,
1376 orig_buf->start, buf->start,
1377 orig_root, ref_root,
1378 orig_generation, ref_generation,
1391 *nr_extents = nr_file_extents;
1393 *nr_extents = nritems;
1402 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root, struct extent_buffer *orig_buf,
1404 struct extent_buffer *buf, int start_slot, int nr)
1411 u64 orig_generation;
1412 struct btrfs_key key;
1413 struct btrfs_file_extent_item *fi;
1419 BUG_ON(start_slot < 0);
1420 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1422 ref_root = btrfs_header_owner(buf);
1423 ref_generation = btrfs_header_generation(buf);
1424 orig_root = btrfs_header_owner(orig_buf);
1425 orig_generation = btrfs_header_generation(orig_buf);
1426 level = btrfs_header_level(buf);
1428 if (!root->ref_cows) {
1430 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1433 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1437 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1440 btrfs_item_key_to_cpu(buf, &key, slot);
1441 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1443 fi = btrfs_item_ptr(buf, slot,
1444 struct btrfs_file_extent_item);
1445 if (btrfs_file_extent_type(buf, fi) ==
1446 BTRFS_FILE_EXTENT_INLINE)
1448 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1451 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1452 btrfs_file_extent_disk_num_bytes(buf, fi),
1453 orig_buf->start, buf->start,
1454 orig_root, ref_root, orig_generation,
1455 ref_generation, key.objectid);
1459 bytenr = btrfs_node_blockptr(buf, slot);
1460 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1461 buf->len, orig_buf->start,
1462 buf->start, orig_root, ref_root,
1463 orig_generation, ref_generation,
1475 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1476 struct btrfs_root *root,
1477 struct btrfs_path *path,
1478 struct btrfs_block_group_cache *cache)
1481 struct btrfs_root *extent_root = root->fs_info->extent_root;
1483 struct extent_buffer *leaf;
1485 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1490 leaf = path->nodes[0];
1491 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1492 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1493 btrfs_mark_buffer_dirty(leaf);
1494 btrfs_release_path(extent_root, path);
1502 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1503 struct btrfs_root *root)
1505 struct btrfs_block_group_cache *cache, *entry;
1509 struct btrfs_path *path;
1512 path = btrfs_alloc_path();
1518 spin_lock(&root->fs_info->block_group_cache_lock);
1519 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1520 n; n = rb_next(n)) {
1521 entry = rb_entry(n, struct btrfs_block_group_cache,
1528 spin_unlock(&root->fs_info->block_group_cache_lock);
1534 last += cache->key.offset;
1536 err = write_one_cache_group(trans, root,
1539 * if we fail to write the cache group, we want
1540 * to keep it marked dirty in hopes that a later
1548 btrfs_free_path(path);
1552 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1554 struct btrfs_block_group_cache *block_group;
1557 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1558 if (!block_group || block_group->ro)
1561 put_block_group(block_group);
1565 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1566 u64 total_bytes, u64 bytes_used,
1567 struct btrfs_space_info **space_info)
1569 struct btrfs_space_info *found;
1571 found = __find_space_info(info, flags);
1573 spin_lock(&found->lock);
1574 found->total_bytes += total_bytes;
1575 found->bytes_used += bytes_used;
1577 spin_unlock(&found->lock);
1578 *space_info = found;
1581 found = kzalloc(sizeof(*found), GFP_NOFS);
1585 INIT_LIST_HEAD(&found->block_groups);
1586 init_rwsem(&found->groups_sem);
1587 spin_lock_init(&found->lock);
1588 found->flags = flags;
1589 found->total_bytes = total_bytes;
1590 found->bytes_used = bytes_used;
1591 found->bytes_pinned = 0;
1592 found->bytes_reserved = 0;
1593 found->bytes_readonly = 0;
1594 found->bytes_delalloc = 0;
1596 found->force_alloc = 0;
1597 *space_info = found;
1598 list_add_rcu(&found->list, &info->space_info);
1602 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1604 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1605 BTRFS_BLOCK_GROUP_RAID1 |
1606 BTRFS_BLOCK_GROUP_RAID10 |
1607 BTRFS_BLOCK_GROUP_DUP);
1609 if (flags & BTRFS_BLOCK_GROUP_DATA)
1610 fs_info->avail_data_alloc_bits |= extra_flags;
1611 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1612 fs_info->avail_metadata_alloc_bits |= extra_flags;
1613 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1614 fs_info->avail_system_alloc_bits |= extra_flags;
1618 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1620 spin_lock(&cache->space_info->lock);
1621 spin_lock(&cache->lock);
1623 cache->space_info->bytes_readonly += cache->key.offset -
1624 btrfs_block_group_used(&cache->item);
1627 spin_unlock(&cache->lock);
1628 spin_unlock(&cache->space_info->lock);
1631 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1633 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1635 if (num_devices == 1)
1636 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1637 if (num_devices < 4)
1638 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1640 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1641 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1642 BTRFS_BLOCK_GROUP_RAID10))) {
1643 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1646 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1647 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1648 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1651 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1652 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1653 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1654 (flags & BTRFS_BLOCK_GROUP_DUP)))
1655 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1659 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1661 struct btrfs_fs_info *info = root->fs_info;
1665 alloc_profile = info->avail_data_alloc_bits &
1666 info->data_alloc_profile;
1667 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1668 } else if (root == root->fs_info->chunk_root) {
1669 alloc_profile = info->avail_system_alloc_bits &
1670 info->system_alloc_profile;
1671 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1673 alloc_profile = info->avail_metadata_alloc_bits &
1674 info->metadata_alloc_profile;
1675 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1678 return btrfs_reduce_alloc_profile(root, data);
1681 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
1685 alloc_target = btrfs_get_alloc_profile(root, 1);
1686 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
1691 * for now this just makes sure we have at least 5% of our metadata space free
1694 int btrfs_check_metadata_free_space(struct btrfs_root *root)
1696 struct btrfs_fs_info *info = root->fs_info;
1697 struct btrfs_space_info *meta_sinfo;
1698 u64 alloc_target, thresh;
1699 int committed = 0, ret;
1701 /* get the space info for where the metadata will live */
1702 alloc_target = btrfs_get_alloc_profile(root, 0);
1703 meta_sinfo = __find_space_info(info, alloc_target);
1706 spin_lock(&meta_sinfo->lock);
1707 if (!meta_sinfo->full)
1708 thresh = meta_sinfo->total_bytes * 80;
1710 thresh = meta_sinfo->total_bytes * 95;
1712 do_div(thresh, 100);
1714 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
1715 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
1716 struct btrfs_trans_handle *trans;
1717 if (!meta_sinfo->full) {
1718 meta_sinfo->force_alloc = 1;
1719 spin_unlock(&meta_sinfo->lock);
1721 trans = btrfs_start_transaction(root, 1);
1725 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1726 2 * 1024 * 1024, alloc_target, 0);
1727 btrfs_end_transaction(trans, root);
1730 spin_unlock(&meta_sinfo->lock);
1734 trans = btrfs_join_transaction(root, 1);
1737 ret = btrfs_commit_transaction(trans, root);
1744 spin_unlock(&meta_sinfo->lock);
1750 * This will check the space that the inode allocates from to make sure we have
1751 * enough space for bytes.
1753 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
1756 struct btrfs_space_info *data_sinfo;
1757 int ret = 0, committed = 0;
1759 /* make sure bytes are sectorsize aligned */
1760 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1762 data_sinfo = BTRFS_I(inode)->space_info;
1764 /* make sure we have enough space to handle the data first */
1765 spin_lock(&data_sinfo->lock);
1766 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
1767 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
1768 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
1769 data_sinfo->bytes_may_use < bytes) {
1770 struct btrfs_trans_handle *trans;
1773 * if we don't have enough free bytes in this space then we need
1774 * to alloc a new chunk.
1776 if (!data_sinfo->full) {
1779 data_sinfo->force_alloc = 1;
1780 spin_unlock(&data_sinfo->lock);
1782 alloc_target = btrfs_get_alloc_profile(root, 1);
1783 trans = btrfs_start_transaction(root, 1);
1787 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1788 bytes + 2 * 1024 * 1024,
1790 btrfs_end_transaction(trans, root);
1795 spin_unlock(&data_sinfo->lock);
1797 /* commit the current transaction and try again */
1800 trans = btrfs_join_transaction(root, 1);
1803 ret = btrfs_commit_transaction(trans, root);
1809 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
1810 ", %llu bytes_used, %llu bytes_reserved, "
1811 "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
1812 "%llu total\n", bytes, data_sinfo->bytes_delalloc,
1813 data_sinfo->bytes_used, data_sinfo->bytes_reserved,
1814 data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
1815 data_sinfo->bytes_may_use, data_sinfo->total_bytes);
1818 data_sinfo->bytes_may_use += bytes;
1819 BTRFS_I(inode)->reserved_bytes += bytes;
1820 spin_unlock(&data_sinfo->lock);
1822 return btrfs_check_metadata_free_space(root);
1826 * if there was an error for whatever reason after calling
1827 * btrfs_check_data_free_space, call this so we can cleanup the counters.
1829 void btrfs_free_reserved_data_space(struct btrfs_root *root,
1830 struct inode *inode, u64 bytes)
1832 struct btrfs_space_info *data_sinfo;
1834 /* make sure bytes are sectorsize aligned */
1835 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1837 data_sinfo = BTRFS_I(inode)->space_info;
1838 spin_lock(&data_sinfo->lock);
1839 data_sinfo->bytes_may_use -= bytes;
1840 BTRFS_I(inode)->reserved_bytes -= bytes;
1841 spin_unlock(&data_sinfo->lock);
1844 /* called when we are adding a delalloc extent to the inode's io_tree */
1845 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
1848 struct btrfs_space_info *data_sinfo;
1850 /* get the space info for where this inode will be storing its data */
1851 data_sinfo = BTRFS_I(inode)->space_info;
1853 /* make sure we have enough space to handle the data first */
1854 spin_lock(&data_sinfo->lock);
1855 data_sinfo->bytes_delalloc += bytes;
1858 * we are adding a delalloc extent without calling
1859 * btrfs_check_data_free_space first. This happens on a weird
1860 * writepage condition, but shouldn't hurt our accounting
1862 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
1863 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
1864 BTRFS_I(inode)->reserved_bytes = 0;
1866 data_sinfo->bytes_may_use -= bytes;
1867 BTRFS_I(inode)->reserved_bytes -= bytes;
1870 spin_unlock(&data_sinfo->lock);
1873 /* called when we are clearing an delalloc extent from the inode's io_tree */
1874 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
1877 struct btrfs_space_info *info;
1879 info = BTRFS_I(inode)->space_info;
1881 spin_lock(&info->lock);
1882 info->bytes_delalloc -= bytes;
1883 spin_unlock(&info->lock);
1886 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1887 struct btrfs_root *extent_root, u64 alloc_bytes,
1888 u64 flags, int force)
1890 struct btrfs_space_info *space_info;
1894 mutex_lock(&extent_root->fs_info->chunk_mutex);
1896 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1898 space_info = __find_space_info(extent_root->fs_info, flags);
1900 ret = update_space_info(extent_root->fs_info, flags,
1904 BUG_ON(!space_info);
1906 spin_lock(&space_info->lock);
1907 if (space_info->force_alloc) {
1909 space_info->force_alloc = 0;
1911 if (space_info->full) {
1912 spin_unlock(&space_info->lock);
1916 thresh = space_info->total_bytes - space_info->bytes_readonly;
1917 thresh = div_factor(thresh, 6);
1919 (space_info->bytes_used + space_info->bytes_pinned +
1920 space_info->bytes_reserved + alloc_bytes) < thresh) {
1921 spin_unlock(&space_info->lock);
1924 spin_unlock(&space_info->lock);
1926 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1928 space_info->full = 1;
1930 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1934 static int update_block_group(struct btrfs_trans_handle *trans,
1935 struct btrfs_root *root,
1936 u64 bytenr, u64 num_bytes, int alloc,
1939 struct btrfs_block_group_cache *cache;
1940 struct btrfs_fs_info *info = root->fs_info;
1941 u64 total = num_bytes;
1946 cache = btrfs_lookup_block_group(info, bytenr);
1949 byte_in_group = bytenr - cache->key.objectid;
1950 WARN_ON(byte_in_group > cache->key.offset);
1952 spin_lock(&cache->space_info->lock);
1953 spin_lock(&cache->lock);
1955 old_val = btrfs_block_group_used(&cache->item);
1956 num_bytes = min(total, cache->key.offset - byte_in_group);
1958 old_val += num_bytes;
1959 cache->space_info->bytes_used += num_bytes;
1961 cache->space_info->bytes_readonly -= num_bytes;
1962 btrfs_set_block_group_used(&cache->item, old_val);
1963 spin_unlock(&cache->lock);
1964 spin_unlock(&cache->space_info->lock);
1966 old_val -= num_bytes;
1967 cache->space_info->bytes_used -= num_bytes;
1969 cache->space_info->bytes_readonly += num_bytes;
1970 btrfs_set_block_group_used(&cache->item, old_val);
1971 spin_unlock(&cache->lock);
1972 spin_unlock(&cache->space_info->lock);
1976 ret = btrfs_discard_extent(root, bytenr,
1980 ret = btrfs_add_free_space(cache, bytenr,
1985 put_block_group(cache);
1987 bytenr += num_bytes;
1992 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1994 struct btrfs_block_group_cache *cache;
1997 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2001 bytenr = cache->key.objectid;
2002 put_block_group(cache);
2007 int btrfs_update_pinned_extents(struct btrfs_root *root,
2008 u64 bytenr, u64 num, int pin)
2011 struct btrfs_block_group_cache *cache;
2012 struct btrfs_fs_info *fs_info = root->fs_info;
2014 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2016 set_extent_dirty(&fs_info->pinned_extents,
2017 bytenr, bytenr + num - 1, GFP_NOFS);
2019 clear_extent_dirty(&fs_info->pinned_extents,
2020 bytenr, bytenr + num - 1, GFP_NOFS);
2023 cache = btrfs_lookup_block_group(fs_info, bytenr);
2025 len = min(num, cache->key.offset -
2026 (bytenr - cache->key.objectid));
2028 spin_lock(&cache->space_info->lock);
2029 spin_lock(&cache->lock);
2030 cache->pinned += len;
2031 cache->space_info->bytes_pinned += len;
2032 spin_unlock(&cache->lock);
2033 spin_unlock(&cache->space_info->lock);
2034 fs_info->total_pinned += len;
2036 spin_lock(&cache->space_info->lock);
2037 spin_lock(&cache->lock);
2038 cache->pinned -= len;
2039 cache->space_info->bytes_pinned -= len;
2040 spin_unlock(&cache->lock);
2041 spin_unlock(&cache->space_info->lock);
2042 fs_info->total_pinned -= len;
2044 btrfs_add_free_space(cache, bytenr, len);
2046 put_block_group(cache);
2053 static int update_reserved_extents(struct btrfs_root *root,
2054 u64 bytenr, u64 num, int reserve)
2057 struct btrfs_block_group_cache *cache;
2058 struct btrfs_fs_info *fs_info = root->fs_info;
2061 cache = btrfs_lookup_block_group(fs_info, bytenr);
2063 len = min(num, cache->key.offset -
2064 (bytenr - cache->key.objectid));
2066 spin_lock(&cache->space_info->lock);
2067 spin_lock(&cache->lock);
2069 cache->reserved += len;
2070 cache->space_info->bytes_reserved += len;
2072 cache->reserved -= len;
2073 cache->space_info->bytes_reserved -= len;
2075 spin_unlock(&cache->lock);
2076 spin_unlock(&cache->space_info->lock);
2077 put_block_group(cache);
2084 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2089 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2092 mutex_lock(&root->fs_info->pinned_mutex);
2094 ret = find_first_extent_bit(pinned_extents, last,
2095 &start, &end, EXTENT_DIRTY);
2098 set_extent_dirty(copy, start, end, GFP_NOFS);
2101 mutex_unlock(&root->fs_info->pinned_mutex);
2105 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *root,
2107 struct extent_io_tree *unpin)
2113 mutex_lock(&root->fs_info->pinned_mutex);
2115 ret = find_first_extent_bit(unpin, 0, &start, &end,
2120 ret = btrfs_discard_extent(root, start, end + 1 - start);
2122 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2123 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2125 if (need_resched()) {
2126 mutex_unlock(&root->fs_info->pinned_mutex);
2128 mutex_lock(&root->fs_info->pinned_mutex);
2131 mutex_unlock(&root->fs_info->pinned_mutex);
2135 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2136 struct btrfs_root *root,
2137 u64 bytenr, u64 num_bytes, int is_data)
2140 struct extent_buffer *buf;
2145 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2149 /* we can reuse a block if it hasn't been written
2150 * and it is from this transaction. We can't
2151 * reuse anything from the tree log root because
2152 * it has tiny sub-transactions.
2154 if (btrfs_buffer_uptodate(buf, 0) &&
2155 btrfs_try_tree_lock(buf)) {
2156 u64 header_owner = btrfs_header_owner(buf);
2157 u64 header_transid = btrfs_header_generation(buf);
2158 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2159 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2160 header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2161 header_transid == trans->transid &&
2162 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2163 clean_tree_block(NULL, root, buf);
2164 btrfs_tree_unlock(buf);
2165 free_extent_buffer(buf);
2168 btrfs_tree_unlock(buf);
2170 free_extent_buffer(buf);
2172 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2179 * remove an extent from the root, returns 0 on success
2181 static int __free_extent(struct btrfs_trans_handle *trans,
2182 struct btrfs_root *root,
2183 u64 bytenr, u64 num_bytes, u64 parent,
2184 u64 root_objectid, u64 ref_generation,
2185 u64 owner_objectid, int pin, int mark_free,
2188 struct btrfs_path *path;
2189 struct btrfs_key key;
2190 struct btrfs_fs_info *info = root->fs_info;
2191 struct btrfs_root *extent_root = info->extent_root;
2192 struct extent_buffer *leaf;
2194 int extent_slot = 0;
2195 int found_extent = 0;
2197 struct btrfs_extent_item *ei;
2200 key.objectid = bytenr;
2201 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2202 key.offset = num_bytes;
2203 path = btrfs_alloc_path();
2208 ret = lookup_extent_backref(trans, extent_root, path,
2209 bytenr, parent, root_objectid,
2210 ref_generation, owner_objectid, 1);
2212 struct btrfs_key found_key;
2213 extent_slot = path->slots[0];
2214 while (extent_slot > 0) {
2216 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2218 if (found_key.objectid != bytenr)
2220 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2221 found_key.offset == num_bytes) {
2225 if (path->slots[0] - extent_slot > 5)
2228 if (!found_extent) {
2229 ret = remove_extent_backref(trans, extent_root, path,
2232 btrfs_release_path(extent_root, path);
2233 ret = btrfs_search_slot(trans, extent_root,
2236 printk(KERN_ERR "umm, got %d back from search"
2237 ", was looking for %llu\n", ret,
2238 (unsigned long long)bytenr);
2239 btrfs_print_leaf(extent_root, path->nodes[0]);
2242 extent_slot = path->slots[0];
2245 btrfs_print_leaf(extent_root, path->nodes[0]);
2247 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2248 "parent %llu root %llu gen %llu owner %llu\n",
2249 (unsigned long long)bytenr,
2250 (unsigned long long)parent,
2251 (unsigned long long)root_objectid,
2252 (unsigned long long)ref_generation,
2253 (unsigned long long)owner_objectid);
2256 leaf = path->nodes[0];
2257 ei = btrfs_item_ptr(leaf, extent_slot,
2258 struct btrfs_extent_item);
2259 refs = btrfs_extent_refs(leaf, ei);
2262 * we're not allowed to delete the extent item if there
2263 * are other delayed ref updates pending
2266 BUG_ON(refs < refs_to_drop);
2267 refs -= refs_to_drop;
2268 btrfs_set_extent_refs(leaf, ei, refs);
2269 btrfs_mark_buffer_dirty(leaf);
2271 if (refs == 0 && found_extent &&
2272 path->slots[0] == extent_slot + 1) {
2273 struct btrfs_extent_ref *ref;
2274 ref = btrfs_item_ptr(leaf, path->slots[0],
2275 struct btrfs_extent_ref);
2276 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2277 /* if the back ref and the extent are next to each other
2278 * they get deleted below in one shot
2280 path->slots[0] = extent_slot;
2282 } else if (found_extent) {
2283 /* otherwise delete the extent back ref */
2284 ret = remove_extent_backref(trans, extent_root, path,
2287 /* if refs are 0, we need to setup the path for deletion */
2289 btrfs_release_path(extent_root, path);
2290 ret = btrfs_search_slot(trans, extent_root, &key, path,
2301 mutex_lock(&root->fs_info->pinned_mutex);
2302 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2303 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2304 mutex_unlock(&root->fs_info->pinned_mutex);
2309 /* block accounting for super block */
2310 spin_lock(&info->delalloc_lock);
2311 super_used = btrfs_super_bytes_used(&info->super_copy);
2312 btrfs_set_super_bytes_used(&info->super_copy,
2313 super_used - num_bytes);
2315 /* block accounting for root item */
2316 root_used = btrfs_root_used(&root->root_item);
2317 btrfs_set_root_used(&root->root_item,
2318 root_used - num_bytes);
2319 spin_unlock(&info->delalloc_lock);
2320 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2323 btrfs_release_path(extent_root, path);
2325 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2326 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2330 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2334 btrfs_free_path(path);
2339 * remove an extent from the root, returns 0 on success
2341 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2342 struct btrfs_root *root,
2343 u64 bytenr, u64 num_bytes, u64 parent,
2344 u64 root_objectid, u64 ref_generation,
2345 u64 owner_objectid, int pin,
2348 WARN_ON(num_bytes < root->sectorsize);
2351 * if metadata always pin
2352 * if data pin when any transaction has committed this
2354 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2355 ref_generation != trans->transid)
2358 if (ref_generation != trans->transid)
2361 return __free_extent(trans, root, bytenr, num_bytes, parent,
2362 root_objectid, ref_generation,
2363 owner_objectid, pin, pin == 0, refs_to_drop);
2367 * when we free an extent, it is possible (and likely) that we free the last
2368 * delayed ref for that extent as well. This searches the delayed ref tree for
2369 * a given extent, and if there are no other delayed refs to be processed, it
2370 * removes it from the tree.
2372 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2373 struct btrfs_root *root, u64 bytenr)
2375 struct btrfs_delayed_ref_head *head;
2376 struct btrfs_delayed_ref_root *delayed_refs;
2377 struct btrfs_delayed_ref_node *ref;
2378 struct rb_node *node;
2381 delayed_refs = &trans->transaction->delayed_refs;
2382 spin_lock(&delayed_refs->lock);
2383 head = btrfs_find_delayed_ref_head(trans, bytenr);
2387 node = rb_prev(&head->node.rb_node);
2391 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2393 /* there are still entries for this ref, we can't drop it */
2394 if (ref->bytenr == bytenr)
2398 * waiting for the lock here would deadlock. If someone else has it
2399 * locked they are already in the process of dropping it anyway
2401 if (!mutex_trylock(&head->mutex))
2405 * at this point we have a head with no other entries. Go
2406 * ahead and process it.
2408 head->node.in_tree = 0;
2409 rb_erase(&head->node.rb_node, &delayed_refs->root);
2410 delayed_refs->num_entries--;
2413 * we don't take a ref on the node because we're removing it from the
2414 * tree, so we just steal the ref the tree was holding.
2416 spin_unlock(&delayed_refs->lock);
2418 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2419 &head->node, head->must_insert_reserved);
2421 btrfs_put_delayed_ref(&head->node);
2424 spin_unlock(&delayed_refs->lock);
2428 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2429 struct btrfs_root *root,
2430 u64 bytenr, u64 num_bytes, u64 parent,
2431 u64 root_objectid, u64 ref_generation,
2432 u64 owner_objectid, int pin)
2437 * tree log blocks never actually go into the extent allocation
2438 * tree, just update pinning info and exit early.
2440 * data extents referenced by the tree log do need to have
2441 * their reference counts bumped.
2443 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2444 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2445 mutex_lock(&root->fs_info->pinned_mutex);
2446 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2447 mutex_unlock(&root->fs_info->pinned_mutex);
2448 update_reserved_extents(root, bytenr, num_bytes, 0);
2451 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2452 root_objectid, ref_generation,
2454 BTRFS_DROP_DELAYED_REF, 1);
2456 ret = check_ref_cleanup(trans, root, bytenr);
2462 static u64 stripe_align(struct btrfs_root *root, u64 val)
2464 u64 mask = ((u64)root->stripesize - 1);
2465 u64 ret = (val + mask) & ~mask;
2470 * walks the btree of allocated extents and find a hole of a given size.
2471 * The key ins is changed to record the hole:
2472 * ins->objectid == block start
2473 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2474 * ins->offset == number of blocks
2475 * Any available blocks before search_start are skipped.
2477 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2478 struct btrfs_root *orig_root,
2479 u64 num_bytes, u64 empty_size,
2480 u64 search_start, u64 search_end,
2481 u64 hint_byte, struct btrfs_key *ins,
2482 u64 exclude_start, u64 exclude_nr,
2486 struct btrfs_root *root = orig_root->fs_info->extent_root;
2487 u64 total_needed = num_bytes;
2488 u64 *last_ptr = NULL;
2489 u64 last_wanted = 0;
2490 struct btrfs_block_group_cache *block_group = NULL;
2491 int chunk_alloc_done = 0;
2492 int empty_cluster = 2 * 1024 * 1024;
2493 int allowed_chunk_alloc = 0;
2494 struct list_head *head = NULL, *cur = NULL;
2497 struct btrfs_space_info *space_info;
2499 WARN_ON(num_bytes < root->sectorsize);
2500 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2504 if (orig_root->ref_cows || empty_size)
2505 allowed_chunk_alloc = 1;
2507 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2508 last_ptr = &root->fs_info->last_alloc;
2509 if (!btrfs_test_opt(root, SSD))
2510 empty_cluster = 64 * 1024;
2513 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2514 last_ptr = &root->fs_info->last_data_alloc;
2518 hint_byte = *last_ptr;
2519 last_wanted = *last_ptr;
2521 empty_size += empty_cluster;
2525 search_start = max(search_start, first_logical_byte(root, 0));
2526 search_start = max(search_start, hint_byte);
2528 if (last_wanted && search_start != last_wanted) {
2530 empty_size += empty_cluster;
2533 total_needed += empty_size;
2534 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2536 block_group = btrfs_lookup_first_block_group(root->fs_info,
2538 space_info = __find_space_info(root->fs_info, data);
2540 down_read(&space_info->groups_sem);
2542 struct btrfs_free_space *free_space;
2544 * the only way this happens if our hint points to a block
2545 * group thats not of the proper type, while looping this
2546 * should never happen
2552 goto new_group_no_lock;
2554 if (unlikely(!block_group->cached)) {
2555 mutex_lock(&block_group->cache_mutex);
2556 ret = cache_block_group(root, block_group);
2557 mutex_unlock(&block_group->cache_mutex);
2562 mutex_lock(&block_group->alloc_mutex);
2563 if (unlikely(!block_group_bits(block_group, data)))
2566 if (unlikely(block_group->ro))
2569 free_space = btrfs_find_free_space(block_group, search_start,
2572 u64 start = block_group->key.objectid;
2573 u64 end = block_group->key.objectid +
2574 block_group->key.offset;
2576 search_start = stripe_align(root, free_space->offset);
2578 /* move on to the next group */
2579 if (search_start + num_bytes >= search_end)
2582 /* move on to the next group */
2583 if (search_start + num_bytes > end)
2586 if (last_wanted && search_start != last_wanted) {
2587 total_needed += empty_cluster;
2588 empty_size += empty_cluster;
2591 * if search_start is still in this block group
2592 * then we just re-search this block group
2594 if (search_start >= start &&
2595 search_start < end) {
2596 mutex_unlock(&block_group->alloc_mutex);
2600 /* else we go to the next block group */
2604 if (exclude_nr > 0 &&
2605 (search_start + num_bytes > exclude_start &&
2606 search_start < exclude_start + exclude_nr)) {
2607 search_start = exclude_start + exclude_nr;
2609 * if search_start is still in this block group
2610 * then we just re-search this block group
2612 if (search_start >= start &&
2613 search_start < end) {
2614 mutex_unlock(&block_group->alloc_mutex);
2619 /* else we go to the next block group */
2623 ins->objectid = search_start;
2624 ins->offset = num_bytes;
2626 btrfs_remove_free_space_lock(block_group, search_start,
2628 /* we are all good, lets return */
2629 mutex_unlock(&block_group->alloc_mutex);
2633 mutex_unlock(&block_group->alloc_mutex);
2634 put_block_group(block_group);
2637 /* don't try to compare new allocations against the
2638 * last allocation any more
2643 * Here's how this works.
2644 * loop == 0: we were searching a block group via a hint
2645 * and didn't find anything, so we start at
2646 * the head of the block groups and keep searching
2647 * loop == 1: we're searching through all of the block groups
2648 * if we hit the head again we have searched
2649 * all of the block groups for this space and we
2650 * need to try and allocate, if we cant error out.
2651 * loop == 2: we allocated more space and are looping through
2652 * all of the block groups again.
2655 head = &space_info->block_groups;
2658 } else if (loop == 1 && cur == head) {
2661 /* at this point we give up on the empty_size
2662 * allocations and just try to allocate the min
2665 * The extra_loop field was set if an empty_size
2666 * allocation was attempted above, and if this
2667 * is try we need to try the loop again without
2668 * the additional empty_size.
2670 total_needed -= empty_size;
2672 keep_going = extra_loop;
2675 if (allowed_chunk_alloc && !chunk_alloc_done) {
2676 up_read(&space_info->groups_sem);
2677 ret = do_chunk_alloc(trans, root, num_bytes +
2678 2 * 1024 * 1024, data, 1);
2679 down_read(&space_info->groups_sem);
2682 head = &space_info->block_groups;
2684 * we've allocated a new chunk, keep
2688 chunk_alloc_done = 1;
2689 } else if (!allowed_chunk_alloc) {
2690 space_info->force_alloc = 1;
2699 } else if (cur == head) {
2703 block_group = list_entry(cur, struct btrfs_block_group_cache,
2705 atomic_inc(&block_group->count);
2707 search_start = block_group->key.objectid;
2711 /* we found what we needed */
2712 if (ins->objectid) {
2713 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2714 trans->block_group = block_group->key.objectid;
2717 *last_ptr = ins->objectid + ins->offset;
2720 printk(KERN_ERR "btrfs searching for %llu bytes, "
2721 "num_bytes %llu, loop %d, allowed_alloc %d\n",
2722 (unsigned long long)total_needed,
2723 (unsigned long long)num_bytes,
2724 loop, allowed_chunk_alloc);
2728 put_block_group(block_group);
2730 up_read(&space_info->groups_sem);
2734 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2736 struct btrfs_block_group_cache *cache;
2738 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
2739 (unsigned long long)(info->total_bytes - info->bytes_used -
2740 info->bytes_pinned - info->bytes_reserved),
2741 (info->full) ? "" : "not ");
2742 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
2743 " may_use=%llu, used=%llu\n", info->total_bytes,
2744 info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
2747 down_read(&info->groups_sem);
2748 list_for_each_entry(cache, &info->block_groups, list) {
2749 spin_lock(&cache->lock);
2750 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
2751 "%llu pinned %llu reserved\n",
2752 (unsigned long long)cache->key.objectid,
2753 (unsigned long long)cache->key.offset,
2754 (unsigned long long)btrfs_block_group_used(&cache->item),
2755 (unsigned long long)cache->pinned,
2756 (unsigned long long)cache->reserved);
2757 btrfs_dump_free_space(cache, bytes);
2758 spin_unlock(&cache->lock);
2760 up_read(&info->groups_sem);
2763 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 u64 num_bytes, u64 min_alloc_size,
2766 u64 empty_size, u64 hint_byte,
2767 u64 search_end, struct btrfs_key *ins,
2771 u64 search_start = 0;
2772 struct btrfs_fs_info *info = root->fs_info;
2774 data = btrfs_get_alloc_profile(root, data);
2777 * the only place that sets empty_size is btrfs_realloc_node, which
2778 * is not called recursively on allocations
2780 if (empty_size || root->ref_cows) {
2781 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2782 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2784 BTRFS_BLOCK_GROUP_METADATA |
2785 (info->metadata_alloc_profile &
2786 info->avail_metadata_alloc_bits), 0);
2788 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2789 num_bytes + 2 * 1024 * 1024, data, 0);
2792 WARN_ON(num_bytes < root->sectorsize);
2793 ret = find_free_extent(trans, root, num_bytes, empty_size,
2794 search_start, search_end, hint_byte, ins,
2795 trans->alloc_exclude_start,
2796 trans->alloc_exclude_nr, data);
2798 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2799 num_bytes = num_bytes >> 1;
2800 num_bytes = num_bytes & ~(root->sectorsize - 1);
2801 num_bytes = max(num_bytes, min_alloc_size);
2802 do_chunk_alloc(trans, root->fs_info->extent_root,
2803 num_bytes, data, 1);
2807 struct btrfs_space_info *sinfo;
2809 sinfo = __find_space_info(root->fs_info, data);
2810 printk(KERN_ERR "btrfs allocation failed flags %llu, "
2811 "wanted %llu\n", (unsigned long long)data,
2812 (unsigned long long)num_bytes);
2813 dump_space_info(sinfo, num_bytes);
2820 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2822 struct btrfs_block_group_cache *cache;
2825 cache = btrfs_lookup_block_group(root->fs_info, start);
2827 printk(KERN_ERR "Unable to find block group for %llu\n",
2828 (unsigned long long)start);
2832 ret = btrfs_discard_extent(root, start, len);
2834 btrfs_add_free_space(cache, start, len);
2835 put_block_group(cache);
2836 update_reserved_extents(root, start, len, 0);
2841 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2842 struct btrfs_root *root,
2843 u64 num_bytes, u64 min_alloc_size,
2844 u64 empty_size, u64 hint_byte,
2845 u64 search_end, struct btrfs_key *ins,
2849 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2850 empty_size, hint_byte, search_end, ins,
2852 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2856 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root, u64 parent,
2858 u64 root_objectid, u64 ref_generation,
2859 u64 owner, struct btrfs_key *ins,
2865 u64 num_bytes = ins->offset;
2867 struct btrfs_fs_info *info = root->fs_info;
2868 struct btrfs_root *extent_root = info->extent_root;
2869 struct btrfs_extent_item *extent_item;
2870 struct btrfs_extent_ref *ref;
2871 struct btrfs_path *path;
2872 struct btrfs_key keys[2];
2875 parent = ins->objectid;
2877 /* block accounting for super block */
2878 spin_lock(&info->delalloc_lock);
2879 super_used = btrfs_super_bytes_used(&info->super_copy);
2880 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2882 /* block accounting for root item */
2883 root_used = btrfs_root_used(&root->root_item);
2884 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2885 spin_unlock(&info->delalloc_lock);
2887 memcpy(&keys[0], ins, sizeof(*ins));
2888 keys[1].objectid = ins->objectid;
2889 keys[1].type = BTRFS_EXTENT_REF_KEY;
2890 keys[1].offset = parent;
2891 sizes[0] = sizeof(*extent_item);
2892 sizes[1] = sizeof(*ref);
2894 path = btrfs_alloc_path();
2897 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2901 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2902 struct btrfs_extent_item);
2903 btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
2904 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2905 struct btrfs_extent_ref);
2907 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2908 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2909 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2910 btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
2912 btrfs_mark_buffer_dirty(path->nodes[0]);
2914 trans->alloc_exclude_start = 0;
2915 trans->alloc_exclude_nr = 0;
2916 btrfs_free_path(path);
2921 ret = update_block_group(trans, root, ins->objectid,
2924 printk(KERN_ERR "btrfs update block group failed for %llu "
2925 "%llu\n", (unsigned long long)ins->objectid,
2926 (unsigned long long)ins->offset);
2933 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2934 struct btrfs_root *root, u64 parent,
2935 u64 root_objectid, u64 ref_generation,
2936 u64 owner, struct btrfs_key *ins)
2940 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2943 ret = btrfs_add_delayed_ref(trans, ins->objectid,
2944 ins->offset, parent, root_objectid,
2945 ref_generation, owner,
2946 BTRFS_ADD_DELAYED_EXTENT, 0);
2952 * this is used by the tree logging recovery code. It records that
2953 * an extent has been allocated and makes sure to clear the free
2954 * space cache bits as well
2956 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2957 struct btrfs_root *root, u64 parent,
2958 u64 root_objectid, u64 ref_generation,
2959 u64 owner, struct btrfs_key *ins)
2962 struct btrfs_block_group_cache *block_group;
2964 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2965 mutex_lock(&block_group->cache_mutex);
2966 cache_block_group(root, block_group);
2967 mutex_unlock(&block_group->cache_mutex);
2969 ret = btrfs_remove_free_space(block_group, ins->objectid,
2972 put_block_group(block_group);
2973 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2974 ref_generation, owner, ins, 1);
2979 * finds a free extent and does all the dirty work required for allocation
2980 * returns the key for the extent through ins, and a tree buffer for
2981 * the first block of the extent through buf.
2983 * returns 0 if everything worked, non-zero otherwise.
2985 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2986 struct btrfs_root *root,
2987 u64 num_bytes, u64 parent, u64 min_alloc_size,
2988 u64 root_objectid, u64 ref_generation,
2989 u64 owner_objectid, u64 empty_size, u64 hint_byte,
2990 u64 search_end, struct btrfs_key *ins, u64 data)
2993 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2994 min_alloc_size, empty_size, hint_byte,
2995 search_end, ins, data);
2997 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2998 ret = btrfs_add_delayed_ref(trans, ins->objectid,
2999 ins->offset, parent, root_objectid,
3000 ref_generation, owner_objectid,
3001 BTRFS_ADD_DELAYED_EXTENT, 0);
3004 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3008 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3009 struct btrfs_root *root,
3010 u64 bytenr, u32 blocksize,
3013 struct extent_buffer *buf;
3015 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3017 return ERR_PTR(-ENOMEM);
3018 btrfs_set_header_generation(buf, trans->transid);
3019 btrfs_set_buffer_lockdep_class(buf, level);
3020 btrfs_tree_lock(buf);
3021 clean_tree_block(trans, root, buf);
3023 btrfs_set_lock_blocking(buf);
3024 btrfs_set_buffer_uptodate(buf);
3026 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3027 set_extent_dirty(&root->dirty_log_pages, buf->start,
3028 buf->start + buf->len - 1, GFP_NOFS);
3030 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3031 buf->start + buf->len - 1, GFP_NOFS);
3033 trans->blocks_used++;
3034 /* this returns a buffer locked for blocking */
3039 * helper function to allocate a block for a given tree
3040 * returns the tree buffer or NULL.
3042 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3043 struct btrfs_root *root,
3044 u32 blocksize, u64 parent,
3051 struct btrfs_key ins;
3053 struct extent_buffer *buf;
3055 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3056 root_objectid, ref_generation, level,
3057 empty_size, hint, (u64)-1, &ins, 0);
3060 return ERR_PTR(ret);
3063 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3068 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3069 struct btrfs_root *root, struct extent_buffer *leaf)
3072 u64 leaf_generation;
3073 struct refsort *sorted;
3074 struct btrfs_key key;
3075 struct btrfs_file_extent_item *fi;
3082 BUG_ON(!btrfs_is_leaf(leaf));
3083 nritems = btrfs_header_nritems(leaf);
3084 leaf_owner = btrfs_header_owner(leaf);
3085 leaf_generation = btrfs_header_generation(leaf);
3087 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3088 /* we do this loop twice. The first time we build a list
3089 * of the extents we have a reference on, then we sort the list
3090 * by bytenr. The second time around we actually do the
3093 for (i = 0; i < nritems; i++) {
3097 btrfs_item_key_to_cpu(leaf, &key, i);
3099 /* only extents have references, skip everything else */
3100 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3103 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3105 /* inline extents live in the btree, they don't have refs */
3106 if (btrfs_file_extent_type(leaf, fi) ==
3107 BTRFS_FILE_EXTENT_INLINE)
3110 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3112 /* holes don't have refs */
3113 if (disk_bytenr == 0)
3116 sorted[refi].bytenr = disk_bytenr;
3117 sorted[refi].slot = i;
3124 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3126 for (i = 0; i < refi; i++) {
3129 disk_bytenr = sorted[i].bytenr;
3130 slot = sorted[i].slot;
3134 btrfs_item_key_to_cpu(leaf, &key, slot);
3135 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3138 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3140 ret = btrfs_free_extent(trans, root, disk_bytenr,
3141 btrfs_file_extent_disk_num_bytes(leaf, fi),
3142 leaf->start, leaf_owner, leaf_generation,
3146 atomic_inc(&root->fs_info->throttle_gen);
3147 wake_up(&root->fs_info->transaction_throttle);
3155 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3156 struct btrfs_root *root,
3157 struct btrfs_leaf_ref *ref)
3161 struct btrfs_extent_info *info;
3162 struct refsort *sorted;
3164 if (ref->nritems == 0)
3167 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3168 for (i = 0; i < ref->nritems; i++) {
3169 sorted[i].bytenr = ref->extents[i].bytenr;
3172 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3175 * the items in the ref were sorted when the ref was inserted
3176 * into the ref cache, so this is already in order
3178 for (i = 0; i < ref->nritems; i++) {
3179 info = ref->extents + sorted[i].slot;
3180 ret = btrfs_free_extent(trans, root, info->bytenr,
3181 info->num_bytes, ref->bytenr,
3182 ref->owner, ref->generation,
3185 atomic_inc(&root->fs_info->throttle_gen);
3186 wake_up(&root->fs_info->transaction_throttle);
3197 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3198 struct btrfs_root *root, u64 start,
3203 ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3206 #if 0 /* some debugging code in case we see problems here */
3207 /* if the refs count is one, it won't get increased again. But
3208 * if the ref count is > 1, someone may be decreasing it at
3209 * the same time we are.
3212 struct extent_buffer *eb = NULL;
3213 eb = btrfs_find_create_tree_block(root, start, len);
3215 btrfs_tree_lock(eb);
3217 mutex_lock(&root->fs_info->alloc_mutex);
3218 ret = lookup_extent_ref(NULL, root, start, len, refs);
3220 mutex_unlock(&root->fs_info->alloc_mutex);
3223 btrfs_tree_unlock(eb);
3224 free_extent_buffer(eb);
3227 printk(KERN_ERR "btrfs block %llu went down to one "
3228 "during drop_snap\n", (unsigned long long)start);
3239 * this is used while deleting old snapshots, and it drops the refs
3240 * on a whole subtree starting from a level 1 node.
3242 * The idea is to sort all the leaf pointers, and then drop the
3243 * ref on all the leaves in order. Most of the time the leaves
3244 * will have ref cache entries, so no leaf IOs will be required to
3245 * find the extents they have references on.
3247 * For each leaf, any references it has are also dropped in order
3249 * This ends up dropping the references in something close to optimal
3250 * order for reading and modifying the extent allocation tree.
3252 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3253 struct btrfs_root *root,
3254 struct btrfs_path *path)
3259 struct extent_buffer *eb = path->nodes[1];
3260 struct extent_buffer *leaf;
3261 struct btrfs_leaf_ref *ref;
3262 struct refsort *sorted = NULL;
3263 int nritems = btrfs_header_nritems(eb);
3267 int slot = path->slots[1];
3268 u32 blocksize = btrfs_level_size(root, 0);
3274 root_owner = btrfs_header_owner(eb);
3275 root_gen = btrfs_header_generation(eb);
3276 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3279 * step one, sort all the leaf pointers so we don't scribble
3280 * randomly into the extent allocation tree
3282 for (i = slot; i < nritems; i++) {
3283 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3284 sorted[refi].slot = i;
3289 * nritems won't be zero, but if we're picking up drop_snapshot
3290 * after a crash, slot might be > 0, so double check things
3296 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3299 * the first loop frees everything the leaves point to
3301 for (i = 0; i < refi; i++) {
3304 bytenr = sorted[i].bytenr;
3307 * check the reference count on this leaf. If it is > 1
3308 * we just decrement it below and don't update any
3309 * of the refs the leaf points to.
3311 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3317 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3320 * the leaf only had one reference, which means the
3321 * only thing pointing to this leaf is the snapshot
3322 * we're deleting. It isn't possible for the reference
3323 * count to increase again later
3325 * The reference cache is checked for the leaf,
3326 * and if found we'll be able to drop any refs held by
3327 * the leaf without needing to read it in.
3329 ref = btrfs_lookup_leaf_ref(root, bytenr);
3330 if (ref && ref->generation != ptr_gen) {
3331 btrfs_free_leaf_ref(root, ref);
3335 ret = cache_drop_leaf_ref(trans, root, ref);
3337 btrfs_remove_leaf_ref(root, ref);
3338 btrfs_free_leaf_ref(root, ref);
3341 * the leaf wasn't in the reference cache, so
3342 * we have to read it.
3344 leaf = read_tree_block(root, bytenr, blocksize,
3346 ret = btrfs_drop_leaf_ref(trans, root, leaf);
3348 free_extent_buffer(leaf);
3350 atomic_inc(&root->fs_info->throttle_gen);
3351 wake_up(&root->fs_info->transaction_throttle);
3356 * run through the loop again to free the refs on the leaves.
3357 * This is faster than doing it in the loop above because
3358 * the leaves are likely to be clustered together. We end up
3359 * working in nice chunks on the extent allocation tree.
3361 for (i = 0; i < refi; i++) {
3362 bytenr = sorted[i].bytenr;
3363 ret = btrfs_free_extent(trans, root, bytenr,
3364 blocksize, eb->start,
3365 root_owner, root_gen, 0, 1);
3368 atomic_inc(&root->fs_info->throttle_gen);
3369 wake_up(&root->fs_info->transaction_throttle);
3376 * update the path to show we've processed the entire level 1
3377 * node. This will get saved into the root's drop_snapshot_progress
3378 * field so these drops are not repeated again if this transaction
3381 path->slots[1] = nritems;
3386 * helper function for drop_snapshot, this walks down the tree dropping ref
3387 * counts as it goes.
3389 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3390 struct btrfs_root *root,
3391 struct btrfs_path *path, int *level)
3397 struct extent_buffer *next;
3398 struct extent_buffer *cur;
3399 struct extent_buffer *parent;
3404 WARN_ON(*level < 0);
3405 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3406 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
3407 path->nodes[*level]->len, &refs);
3413 * walk down to the last node level and free all the leaves
3415 while (*level >= 0) {
3416 WARN_ON(*level < 0);
3417 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3418 cur = path->nodes[*level];
3420 if (btrfs_header_level(cur) != *level)
3423 if (path->slots[*level] >=
3424 btrfs_header_nritems(cur))
3427 /* the new code goes down to level 1 and does all the
3428 * leaves pointed to that node in bulk. So, this check
3429 * for level 0 will always be false.
3431 * But, the disk format allows the drop_snapshot_progress
3432 * field in the root to leave things in a state where
3433 * a leaf will need cleaning up here. If someone crashes
3434 * with the old code and then boots with the new code,
3435 * we might find a leaf here.
3438 ret = btrfs_drop_leaf_ref(trans, root, cur);
3444 * once we get to level one, process the whole node
3445 * at once, including everything below it.
3448 ret = drop_level_one_refs(trans, root, path);
3453 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3454 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3455 blocksize = btrfs_level_size(root, *level - 1);
3457 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3462 * if there is more than one reference, we don't need
3463 * to read that node to drop any references it has. We
3464 * just drop the ref we hold on that node and move on to the
3465 * next slot in this level.
3468 parent = path->nodes[*level];
3469 root_owner = btrfs_header_owner(parent);
3470 root_gen = btrfs_header_generation(parent);
3471 path->slots[*level]++;
3473 ret = btrfs_free_extent(trans, root, bytenr,
3474 blocksize, parent->start,
3475 root_owner, root_gen,
3479 atomic_inc(&root->fs_info->throttle_gen);
3480 wake_up(&root->fs_info->transaction_throttle);
3487 * we need to keep freeing things in the next level down.
3488 * read the block and loop around to process it
3490 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3491 WARN_ON(*level <= 0);
3492 if (path->nodes[*level-1])
3493 free_extent_buffer(path->nodes[*level-1]);
3494 path->nodes[*level-1] = next;
3495 *level = btrfs_header_level(next);
3496 path->slots[*level] = 0;
3500 WARN_ON(*level < 0);
3501 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3503 if (path->nodes[*level] == root->node) {
3504 parent = path->nodes[*level];
3505 bytenr = path->nodes[*level]->start;
3507 parent = path->nodes[*level + 1];
3508 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3511 blocksize = btrfs_level_size(root, *level);
3512 root_owner = btrfs_header_owner(parent);
3513 root_gen = btrfs_header_generation(parent);
3516 * cleanup and free the reference on the last node
3519 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3520 parent->start, root_owner, root_gen,
3522 free_extent_buffer(path->nodes[*level]);
3523 path->nodes[*level] = NULL;
3533 * helper function for drop_subtree, this function is similar to
3534 * walk_down_tree. The main difference is that it checks reference
3535 * counts while tree blocks are locked.
3537 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3538 struct btrfs_root *root,
3539 struct btrfs_path *path, int *level)
3541 struct extent_buffer *next;
3542 struct extent_buffer *cur;
3543 struct extent_buffer *parent;
3550 cur = path->nodes[*level];
3551 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3557 while (*level >= 0) {
3558 cur = path->nodes[*level];
3560 ret = btrfs_drop_leaf_ref(trans, root, cur);
3562 clean_tree_block(trans, root, cur);
3565 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3566 clean_tree_block(trans, root, cur);
3570 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3571 blocksize = btrfs_level_size(root, *level - 1);
3572 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3574 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3575 btrfs_tree_lock(next);
3576 btrfs_set_lock_blocking(next);
3578 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3582 parent = path->nodes[*level];
3583 ret = btrfs_free_extent(trans, root, bytenr,
3584 blocksize, parent->start,
3585 btrfs_header_owner(parent),
3586 btrfs_header_generation(parent),
3589 path->slots[*level]++;
3590 btrfs_tree_unlock(next);
3591 free_extent_buffer(next);
3595 *level = btrfs_header_level(next);
3596 path->nodes[*level] = next;
3597 path->slots[*level] = 0;
3598 path->locks[*level] = 1;
3602 parent = path->nodes[*level + 1];
3603 bytenr = path->nodes[*level]->start;
3604 blocksize = path->nodes[*level]->len;
3606 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3607 parent->start, btrfs_header_owner(parent),
3608 btrfs_header_generation(parent), *level, 1);
3611 if (path->locks[*level]) {
3612 btrfs_tree_unlock(path->nodes[*level]);
3613 path->locks[*level] = 0;
3615 free_extent_buffer(path->nodes[*level]);
3616 path->nodes[*level] = NULL;
3623 * helper for dropping snapshots. This walks back up the tree in the path
3624 * to find the first node higher up where we haven't yet gone through
3627 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3628 struct btrfs_root *root,
3629 struct btrfs_path *path,
3630 int *level, int max_level)
3634 struct btrfs_root_item *root_item = &root->root_item;
3639 for (i = *level; i < max_level && path->nodes[i]; i++) {
3640 slot = path->slots[i];
3641 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3642 struct extent_buffer *node;
3643 struct btrfs_disk_key disk_key;
3646 * there is more work to do in this level.
3647 * Update the drop_progress marker to reflect
3648 * the work we've done so far, and then bump
3651 node = path->nodes[i];
3654 WARN_ON(*level == 0);
3655 btrfs_node_key(node, &disk_key, path->slots[i]);
3656 memcpy(&root_item->drop_progress,
3657 &disk_key, sizeof(disk_key));
3658 root_item->drop_level = i;
3661 struct extent_buffer *parent;
3664 * this whole node is done, free our reference
3665 * on it and go up one level
3667 if (path->nodes[*level] == root->node)
3668 parent = path->nodes[*level];
3670 parent = path->nodes[*level + 1];
3672 root_owner = btrfs_header_owner(parent);
3673 root_gen = btrfs_header_generation(parent);
3675 clean_tree_block(trans, root, path->nodes[*level]);
3676 ret = btrfs_free_extent(trans, root,
3677 path->nodes[*level]->start,
3678 path->nodes[*level]->len,
3679 parent->start, root_owner,
3680 root_gen, *level, 1);
3682 if (path->locks[*level]) {
3683 btrfs_tree_unlock(path->nodes[*level]);
3684 path->locks[*level] = 0;
3686 free_extent_buffer(path->nodes[*level]);
3687 path->nodes[*level] = NULL;
3695 * drop the reference count on the tree rooted at 'snap'. This traverses
3696 * the tree freeing any blocks that have a ref count of zero after being
3699 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3705 struct btrfs_path *path;
3708 struct btrfs_root_item *root_item = &root->root_item;
3710 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3711 path = btrfs_alloc_path();
3714 level = btrfs_header_level(root->node);
3716 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3717 path->nodes[level] = root->node;
3718 extent_buffer_get(root->node);
3719 path->slots[level] = 0;
3721 struct btrfs_key key;
3722 struct btrfs_disk_key found_key;
3723 struct extent_buffer *node;
3725 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3726 level = root_item->drop_level;
3727 path->lowest_level = level;
3728 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3733 node = path->nodes[level];
3734 btrfs_node_key(node, &found_key, path->slots[level]);
3735 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3736 sizeof(found_key)));
3738 * unlock our path, this is safe because only this
3739 * function is allowed to delete this snapshot
3741 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3742 if (path->nodes[i] && path->locks[i]) {
3744 btrfs_tree_unlock(path->nodes[i]);
3749 wret = walk_down_tree(trans, root, path, &level);
3755 wret = walk_up_tree(trans, root, path, &level,
3761 if (trans->transaction->in_commit) {
3765 atomic_inc(&root->fs_info->throttle_gen);
3766 wake_up(&root->fs_info->transaction_throttle);
3768 for (i = 0; i <= orig_level; i++) {
3769 if (path->nodes[i]) {
3770 free_extent_buffer(path->nodes[i]);
3771 path->nodes[i] = NULL;
3775 btrfs_free_path(path);
3779 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3780 struct btrfs_root *root,
3781 struct extent_buffer *node,
3782 struct extent_buffer *parent)
3784 struct btrfs_path *path;
3790 path = btrfs_alloc_path();
3793 btrfs_assert_tree_locked(parent);
3794 parent_level = btrfs_header_level(parent);
3795 extent_buffer_get(parent);
3796 path->nodes[parent_level] = parent;
3797 path->slots[parent_level] = btrfs_header_nritems(parent);
3799 btrfs_assert_tree_locked(node);
3800 level = btrfs_header_level(node);
3801 extent_buffer_get(node);
3802 path->nodes[level] = node;
3803 path->slots[level] = 0;
3806 wret = walk_down_subtree(trans, root, path, &level);
3812 wret = walk_up_tree(trans, root, path, &level, parent_level);
3819 btrfs_free_path(path);
3823 static unsigned long calc_ra(unsigned long start, unsigned long last,
3826 return min(last, start + nr - 1);
3829 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3834 unsigned long first_index;
3835 unsigned long last_index;
3838 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3839 struct file_ra_state *ra;
3840 struct btrfs_ordered_extent *ordered;
3841 unsigned int total_read = 0;
3842 unsigned int total_dirty = 0;
3845 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3847 mutex_lock(&inode->i_mutex);
3848 first_index = start >> PAGE_CACHE_SHIFT;
3849 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3851 /* make sure the dirty trick played by the caller work */
3852 ret = invalidate_inode_pages2_range(inode->i_mapping,
3853 first_index, last_index);
3857 file_ra_state_init(ra, inode->i_mapping);
3859 for (i = first_index ; i <= last_index; i++) {
3860 if (total_read % ra->ra_pages == 0) {
3861 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3862 calc_ra(i, last_index, ra->ra_pages));
3866 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3868 page = grab_cache_page(inode->i_mapping, i);
3873 if (!PageUptodate(page)) {
3874 btrfs_readpage(NULL, page);
3876 if (!PageUptodate(page)) {
3878 page_cache_release(page);
3883 wait_on_page_writeback(page);
3885 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3886 page_end = page_start + PAGE_CACHE_SIZE - 1;
3887 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3889 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3891 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3893 page_cache_release(page);
3894 btrfs_start_ordered_extent(inode, ordered, 1);
3895 btrfs_put_ordered_extent(ordered);
3898 set_page_extent_mapped(page);
3900 if (i == first_index)
3901 set_extent_bits(io_tree, page_start, page_end,
3902 EXTENT_BOUNDARY, GFP_NOFS);
3903 btrfs_set_extent_delalloc(inode, page_start, page_end);
3905 set_page_dirty(page);
3908 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3910 page_cache_release(page);
3915 mutex_unlock(&inode->i_mutex);
3916 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3920 static noinline int relocate_data_extent(struct inode *reloc_inode,
3921 struct btrfs_key *extent_key,
3924 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3925 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3926 struct extent_map *em;
3927 u64 start = extent_key->objectid - offset;
3928 u64 end = start + extent_key->offset - 1;
3930 em = alloc_extent_map(GFP_NOFS);
3931 BUG_ON(!em || IS_ERR(em));
3934 em->len = extent_key->offset;
3935 em->block_len = extent_key->offset;
3936 em->block_start = extent_key->objectid;
3937 em->bdev = root->fs_info->fs_devices->latest_bdev;
3938 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3940 /* setup extent map to cheat btrfs_readpage */
3941 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3944 spin_lock(&em_tree->lock);
3945 ret = add_extent_mapping(em_tree, em);
3946 spin_unlock(&em_tree->lock);
3947 if (ret != -EEXIST) {
3948 free_extent_map(em);
3951 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3953 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3955 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3958 struct btrfs_ref_path {
3960 u64 nodes[BTRFS_MAX_LEVEL];
3962 u64 root_generation;
3969 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3970 u64 new_nodes[BTRFS_MAX_LEVEL];
3973 struct disk_extent {
3984 static int is_cowonly_root(u64 root_objectid)
3986 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3987 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3988 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3989 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3990 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
3991 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
3996 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
3997 struct btrfs_root *extent_root,
3998 struct btrfs_ref_path *ref_path,
4001 struct extent_buffer *leaf;
4002 struct btrfs_path *path;
4003 struct btrfs_extent_ref *ref;
4004 struct btrfs_key key;
4005 struct btrfs_key found_key;
4011 path = btrfs_alloc_path();
4016 ref_path->lowest_level = -1;
4017 ref_path->current_level = -1;
4018 ref_path->shared_level = -1;
4022 level = ref_path->current_level - 1;
4023 while (level >= -1) {
4025 if (level < ref_path->lowest_level)
4029 bytenr = ref_path->nodes[level];
4031 bytenr = ref_path->extent_start;
4032 BUG_ON(bytenr == 0);
4034 parent = ref_path->nodes[level + 1];
4035 ref_path->nodes[level + 1] = 0;
4036 ref_path->current_level = level;
4037 BUG_ON(parent == 0);
4039 key.objectid = bytenr;
4040 key.offset = parent + 1;
4041 key.type = BTRFS_EXTENT_REF_KEY;
4043 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4048 leaf = path->nodes[0];
4049 nritems = btrfs_header_nritems(leaf);
4050 if (path->slots[0] >= nritems) {
4051 ret = btrfs_next_leaf(extent_root, path);
4056 leaf = path->nodes[0];
4059 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4060 if (found_key.objectid == bytenr &&
4061 found_key.type == BTRFS_EXTENT_REF_KEY) {
4062 if (level < ref_path->shared_level)
4063 ref_path->shared_level = level;
4068 btrfs_release_path(extent_root, path);
4071 /* reached lowest level */
4075 level = ref_path->current_level;
4076 while (level < BTRFS_MAX_LEVEL - 1) {
4080 bytenr = ref_path->nodes[level];
4082 bytenr = ref_path->extent_start;
4084 BUG_ON(bytenr == 0);
4086 key.objectid = bytenr;
4088 key.type = BTRFS_EXTENT_REF_KEY;
4090 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4094 leaf = path->nodes[0];
4095 nritems = btrfs_header_nritems(leaf);
4096 if (path->slots[0] >= nritems) {
4097 ret = btrfs_next_leaf(extent_root, path);
4101 /* the extent was freed by someone */
4102 if (ref_path->lowest_level == level)
4104 btrfs_release_path(extent_root, path);
4107 leaf = path->nodes[0];
4110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4111 if (found_key.objectid != bytenr ||
4112 found_key.type != BTRFS_EXTENT_REF_KEY) {
4113 /* the extent was freed by someone */
4114 if (ref_path->lowest_level == level) {
4118 btrfs_release_path(extent_root, path);
4122 ref = btrfs_item_ptr(leaf, path->slots[0],
4123 struct btrfs_extent_ref);
4124 ref_objectid = btrfs_ref_objectid(leaf, ref);
4125 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4127 level = (int)ref_objectid;
4128 BUG_ON(level >= BTRFS_MAX_LEVEL);
4129 ref_path->lowest_level = level;
4130 ref_path->current_level = level;
4131 ref_path->nodes[level] = bytenr;
4133 WARN_ON(ref_objectid != level);
4136 WARN_ON(level != -1);
4140 if (ref_path->lowest_level == level) {
4141 ref_path->owner_objectid = ref_objectid;
4142 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4146 * the block is tree root or the block isn't in reference
4149 if (found_key.objectid == found_key.offset ||
4150 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4151 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4152 ref_path->root_generation =
4153 btrfs_ref_generation(leaf, ref);
4155 /* special reference from the tree log */
4156 ref_path->nodes[0] = found_key.offset;
4157 ref_path->current_level = 0;
4164 BUG_ON(ref_path->nodes[level] != 0);
4165 ref_path->nodes[level] = found_key.offset;
4166 ref_path->current_level = level;
4169 * the reference was created in the running transaction,
4170 * no need to continue walking up.
4172 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4173 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4174 ref_path->root_generation =
4175 btrfs_ref_generation(leaf, ref);
4180 btrfs_release_path(extent_root, path);
4183 /* reached max tree level, but no tree root found. */
4186 btrfs_free_path(path);
4190 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4191 struct btrfs_root *extent_root,
4192 struct btrfs_ref_path *ref_path,
4195 memset(ref_path, 0, sizeof(*ref_path));
4196 ref_path->extent_start = extent_start;
4198 return __next_ref_path(trans, extent_root, ref_path, 1);
4201 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4202 struct btrfs_root *extent_root,
4203 struct btrfs_ref_path *ref_path)
4205 return __next_ref_path(trans, extent_root, ref_path, 0);
4208 static noinline int get_new_locations(struct inode *reloc_inode,
4209 struct btrfs_key *extent_key,
4210 u64 offset, int no_fragment,
4211 struct disk_extent **extents,
4214 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4215 struct btrfs_path *path;
4216 struct btrfs_file_extent_item *fi;
4217 struct extent_buffer *leaf;
4218 struct disk_extent *exts = *extents;
4219 struct btrfs_key found_key;
4224 int max = *nr_extents;
4227 WARN_ON(!no_fragment && *extents);
4230 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4235 path = btrfs_alloc_path();
4238 cur_pos = extent_key->objectid - offset;
4239 last_byte = extent_key->objectid + extent_key->offset;
4240 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4250 leaf = path->nodes[0];
4251 nritems = btrfs_header_nritems(leaf);
4252 if (path->slots[0] >= nritems) {
4253 ret = btrfs_next_leaf(root, path);
4258 leaf = path->nodes[0];
4261 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4262 if (found_key.offset != cur_pos ||
4263 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4264 found_key.objectid != reloc_inode->i_ino)
4267 fi = btrfs_item_ptr(leaf, path->slots[0],
4268 struct btrfs_file_extent_item);
4269 if (btrfs_file_extent_type(leaf, fi) !=
4270 BTRFS_FILE_EXTENT_REG ||
4271 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4275 struct disk_extent *old = exts;
4277 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4278 memcpy(exts, old, sizeof(*exts) * nr);
4279 if (old != *extents)
4283 exts[nr].disk_bytenr =
4284 btrfs_file_extent_disk_bytenr(leaf, fi);
4285 exts[nr].disk_num_bytes =
4286 btrfs_file_extent_disk_num_bytes(leaf, fi);
4287 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4288 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4289 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4290 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4291 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4292 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4294 BUG_ON(exts[nr].offset > 0);
4295 BUG_ON(exts[nr].compression || exts[nr].encryption);
4296 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4298 cur_pos += exts[nr].num_bytes;
4301 if (cur_pos + offset >= last_byte)
4311 BUG_ON(cur_pos + offset > last_byte);
4312 if (cur_pos + offset < last_byte) {
4318 btrfs_free_path(path);
4320 if (exts != *extents)
4329 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4330 struct btrfs_root *root,
4331 struct btrfs_path *path,
4332 struct btrfs_key *extent_key,
4333 struct btrfs_key *leaf_key,
4334 struct btrfs_ref_path *ref_path,
4335 struct disk_extent *new_extents,
4338 struct extent_buffer *leaf;
4339 struct btrfs_file_extent_item *fi;
4340 struct inode *inode = NULL;
4341 struct btrfs_key key;
4346 u64 search_end = (u64)-1;
4349 int extent_locked = 0;
4353 memcpy(&key, leaf_key, sizeof(key));
4354 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4355 if (key.objectid < ref_path->owner_objectid ||
4356 (key.objectid == ref_path->owner_objectid &&
4357 key.type < BTRFS_EXTENT_DATA_KEY)) {
4358 key.objectid = ref_path->owner_objectid;
4359 key.type = BTRFS_EXTENT_DATA_KEY;
4365 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4369 leaf = path->nodes[0];
4370 nritems = btrfs_header_nritems(leaf);
4372 if (extent_locked && ret > 0) {
4374 * the file extent item was modified by someone
4375 * before the extent got locked.
4377 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4378 lock_end, GFP_NOFS);
4382 if (path->slots[0] >= nritems) {
4383 if (++nr_scaned > 2)
4386 BUG_ON(extent_locked);
4387 ret = btrfs_next_leaf(root, path);
4392 leaf = path->nodes[0];
4393 nritems = btrfs_header_nritems(leaf);
4396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4398 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4399 if ((key.objectid > ref_path->owner_objectid) ||
4400 (key.objectid == ref_path->owner_objectid &&
4401 key.type > BTRFS_EXTENT_DATA_KEY) ||
4402 key.offset >= search_end)
4406 if (inode && key.objectid != inode->i_ino) {
4407 BUG_ON(extent_locked);
4408 btrfs_release_path(root, path);
4409 mutex_unlock(&inode->i_mutex);
4415 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4420 fi = btrfs_item_ptr(leaf, path->slots[0],
4421 struct btrfs_file_extent_item);
4422 extent_type = btrfs_file_extent_type(leaf, fi);
4423 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4424 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4425 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4426 extent_key->objectid)) {
4432 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4433 ext_offset = btrfs_file_extent_offset(leaf, fi);
4435 if (search_end == (u64)-1) {
4436 search_end = key.offset - ext_offset +
4437 btrfs_file_extent_ram_bytes(leaf, fi);
4440 if (!extent_locked) {
4441 lock_start = key.offset;
4442 lock_end = lock_start + num_bytes - 1;
4444 if (lock_start > key.offset ||
4445 lock_end + 1 < key.offset + num_bytes) {
4446 unlock_extent(&BTRFS_I(inode)->io_tree,
4447 lock_start, lock_end, GFP_NOFS);
4453 btrfs_release_path(root, path);
4455 inode = btrfs_iget_locked(root->fs_info->sb,
4456 key.objectid, root);
4457 if (inode->i_state & I_NEW) {
4458 BTRFS_I(inode)->root = root;
4459 BTRFS_I(inode)->location.objectid =
4461 BTRFS_I(inode)->location.type =
4462 BTRFS_INODE_ITEM_KEY;
4463 BTRFS_I(inode)->location.offset = 0;
4464 btrfs_read_locked_inode(inode);
4465 unlock_new_inode(inode);
4468 * some code call btrfs_commit_transaction while
4469 * holding the i_mutex, so we can't use mutex_lock
4472 if (is_bad_inode(inode) ||
4473 !mutex_trylock(&inode->i_mutex)) {
4476 key.offset = (u64)-1;
4481 if (!extent_locked) {
4482 struct btrfs_ordered_extent *ordered;
4484 btrfs_release_path(root, path);
4486 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4487 lock_end, GFP_NOFS);
4488 ordered = btrfs_lookup_first_ordered_extent(inode,
4491 ordered->file_offset <= lock_end &&
4492 ordered->file_offset + ordered->len > lock_start) {
4493 unlock_extent(&BTRFS_I(inode)->io_tree,
4494 lock_start, lock_end, GFP_NOFS);
4495 btrfs_start_ordered_extent(inode, ordered, 1);
4496 btrfs_put_ordered_extent(ordered);
4497 key.offset += num_bytes;
4501 btrfs_put_ordered_extent(ordered);
4507 if (nr_extents == 1) {
4508 /* update extent pointer in place */
4509 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4510 new_extents[0].disk_bytenr);
4511 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4512 new_extents[0].disk_num_bytes);
4513 btrfs_mark_buffer_dirty(leaf);
4515 btrfs_drop_extent_cache(inode, key.offset,
4516 key.offset + num_bytes - 1, 0);
4518 ret = btrfs_inc_extent_ref(trans, root,
4519 new_extents[0].disk_bytenr,
4520 new_extents[0].disk_num_bytes,
4522 root->root_key.objectid,
4527 ret = btrfs_free_extent(trans, root,
4528 extent_key->objectid,
4531 btrfs_header_owner(leaf),
4532 btrfs_header_generation(leaf),
4536 btrfs_release_path(root, path);
4537 key.offset += num_bytes;
4545 * drop old extent pointer at first, then insert the
4546 * new pointers one bye one
4548 btrfs_release_path(root, path);
4549 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4550 key.offset + num_bytes,
4551 key.offset, &alloc_hint);
4554 for (i = 0; i < nr_extents; i++) {
4555 if (ext_offset >= new_extents[i].num_bytes) {
4556 ext_offset -= new_extents[i].num_bytes;
4559 extent_len = min(new_extents[i].num_bytes -
4560 ext_offset, num_bytes);
4562 ret = btrfs_insert_empty_item(trans, root,
4567 leaf = path->nodes[0];
4568 fi = btrfs_item_ptr(leaf, path->slots[0],
4569 struct btrfs_file_extent_item);
4570 btrfs_set_file_extent_generation(leaf, fi,
4572 btrfs_set_file_extent_type(leaf, fi,
4573 BTRFS_FILE_EXTENT_REG);
4574 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4575 new_extents[i].disk_bytenr);
4576 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4577 new_extents[i].disk_num_bytes);
4578 btrfs_set_file_extent_ram_bytes(leaf, fi,
4579 new_extents[i].ram_bytes);
4581 btrfs_set_file_extent_compression(leaf, fi,
4582 new_extents[i].compression);
4583 btrfs_set_file_extent_encryption(leaf, fi,
4584 new_extents[i].encryption);
4585 btrfs_set_file_extent_other_encoding(leaf, fi,
4586 new_extents[i].other_encoding);
4588 btrfs_set_file_extent_num_bytes(leaf, fi,
4590 ext_offset += new_extents[i].offset;
4591 btrfs_set_file_extent_offset(leaf, fi,
4593 btrfs_mark_buffer_dirty(leaf);
4595 btrfs_drop_extent_cache(inode, key.offset,
4596 key.offset + extent_len - 1, 0);
4598 ret = btrfs_inc_extent_ref(trans, root,
4599 new_extents[i].disk_bytenr,
4600 new_extents[i].disk_num_bytes,
4602 root->root_key.objectid,
4603 trans->transid, key.objectid);
4605 btrfs_release_path(root, path);
4607 inode_add_bytes(inode, extent_len);
4610 num_bytes -= extent_len;
4611 key.offset += extent_len;
4616 BUG_ON(i >= nr_extents);
4620 if (extent_locked) {
4621 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4622 lock_end, GFP_NOFS);
4626 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4627 key.offset >= search_end)
4634 btrfs_release_path(root, path);
4636 mutex_unlock(&inode->i_mutex);
4637 if (extent_locked) {
4638 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4639 lock_end, GFP_NOFS);
4646 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4647 struct btrfs_root *root,
4648 struct extent_buffer *buf, u64 orig_start)
4653 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4654 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4656 level = btrfs_header_level(buf);
4658 struct btrfs_leaf_ref *ref;
4659 struct btrfs_leaf_ref *orig_ref;
4661 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4665 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4667 btrfs_free_leaf_ref(root, orig_ref);
4671 ref->nritems = orig_ref->nritems;
4672 memcpy(ref->extents, orig_ref->extents,
4673 sizeof(ref->extents[0]) * ref->nritems);
4675 btrfs_free_leaf_ref(root, orig_ref);
4677 ref->root_gen = trans->transid;
4678 ref->bytenr = buf->start;
4679 ref->owner = btrfs_header_owner(buf);
4680 ref->generation = btrfs_header_generation(buf);
4682 ret = btrfs_add_leaf_ref(root, ref, 0);
4684 btrfs_free_leaf_ref(root, ref);
4689 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4690 struct extent_buffer *leaf,
4691 struct btrfs_block_group_cache *group,
4692 struct btrfs_root *target_root)
4694 struct btrfs_key key;
4695 struct inode *inode = NULL;
4696 struct btrfs_file_extent_item *fi;
4698 u64 skip_objectid = 0;
4702 nritems = btrfs_header_nritems(leaf);
4703 for (i = 0; i < nritems; i++) {
4704 btrfs_item_key_to_cpu(leaf, &key, i);
4705 if (key.objectid == skip_objectid ||
4706 key.type != BTRFS_EXTENT_DATA_KEY)
4708 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4709 if (btrfs_file_extent_type(leaf, fi) ==
4710 BTRFS_FILE_EXTENT_INLINE)
4712 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4714 if (!inode || inode->i_ino != key.objectid) {
4716 inode = btrfs_ilookup(target_root->fs_info->sb,
4717 key.objectid, target_root, 1);
4720 skip_objectid = key.objectid;
4723 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4725 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4726 key.offset + num_bytes - 1, GFP_NOFS);
4727 btrfs_drop_extent_cache(inode, key.offset,
4728 key.offset + num_bytes - 1, 1);
4729 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4730 key.offset + num_bytes - 1, GFP_NOFS);
4737 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4738 struct btrfs_root *root,
4739 struct extent_buffer *leaf,
4740 struct btrfs_block_group_cache *group,
4741 struct inode *reloc_inode)
4743 struct btrfs_key key;
4744 struct btrfs_key extent_key;
4745 struct btrfs_file_extent_item *fi;
4746 struct btrfs_leaf_ref *ref;
4747 struct disk_extent *new_extent;
4756 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4757 BUG_ON(!new_extent);
4759 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4763 nritems = btrfs_header_nritems(leaf);
4764 for (i = 0; i < nritems; i++) {
4765 btrfs_item_key_to_cpu(leaf, &key, i);
4766 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4768 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4769 if (btrfs_file_extent_type(leaf, fi) ==
4770 BTRFS_FILE_EXTENT_INLINE)
4772 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4773 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4778 if (bytenr >= group->key.objectid + group->key.offset ||
4779 bytenr + num_bytes <= group->key.objectid)
4782 extent_key.objectid = bytenr;
4783 extent_key.offset = num_bytes;
4784 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4786 ret = get_new_locations(reloc_inode, &extent_key,
4787 group->key.objectid, 1,
4788 &new_extent, &nr_extent);
4793 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4794 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4795 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4796 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4798 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4799 new_extent->disk_bytenr);
4800 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4801 new_extent->disk_num_bytes);
4802 btrfs_mark_buffer_dirty(leaf);
4804 ret = btrfs_inc_extent_ref(trans, root,
4805 new_extent->disk_bytenr,
4806 new_extent->disk_num_bytes,
4808 root->root_key.objectid,
4809 trans->transid, key.objectid);
4812 ret = btrfs_free_extent(trans, root,
4813 bytenr, num_bytes, leaf->start,
4814 btrfs_header_owner(leaf),
4815 btrfs_header_generation(leaf),
4821 BUG_ON(ext_index + 1 != ref->nritems);
4822 btrfs_free_leaf_ref(root, ref);
4826 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4827 struct btrfs_root *root)
4829 struct btrfs_root *reloc_root;
4832 if (root->reloc_root) {
4833 reloc_root = root->reloc_root;
4834 root->reloc_root = NULL;
4835 list_add(&reloc_root->dead_list,
4836 &root->fs_info->dead_reloc_roots);
4838 btrfs_set_root_bytenr(&reloc_root->root_item,
4839 reloc_root->node->start);
4840 btrfs_set_root_level(&root->root_item,
4841 btrfs_header_level(reloc_root->node));
4842 memset(&reloc_root->root_item.drop_progress, 0,
4843 sizeof(struct btrfs_disk_key));
4844 reloc_root->root_item.drop_level = 0;
4846 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4847 &reloc_root->root_key,
4848 &reloc_root->root_item);
4854 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4856 struct btrfs_trans_handle *trans;
4857 struct btrfs_root *reloc_root;
4858 struct btrfs_root *prev_root = NULL;
4859 struct list_head dead_roots;
4863 INIT_LIST_HEAD(&dead_roots);
4864 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4866 while (!list_empty(&dead_roots)) {
4867 reloc_root = list_entry(dead_roots.prev,
4868 struct btrfs_root, dead_list);
4869 list_del_init(&reloc_root->dead_list);
4871 BUG_ON(reloc_root->commit_root != NULL);
4873 trans = btrfs_join_transaction(root, 1);
4876 mutex_lock(&root->fs_info->drop_mutex);
4877 ret = btrfs_drop_snapshot(trans, reloc_root);
4880 mutex_unlock(&root->fs_info->drop_mutex);
4882 nr = trans->blocks_used;
4883 ret = btrfs_end_transaction(trans, root);
4885 btrfs_btree_balance_dirty(root, nr);
4888 free_extent_buffer(reloc_root->node);
4890 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4891 &reloc_root->root_key);
4893 mutex_unlock(&root->fs_info->drop_mutex);
4895 nr = trans->blocks_used;
4896 ret = btrfs_end_transaction(trans, root);
4898 btrfs_btree_balance_dirty(root, nr);
4901 prev_root = reloc_root;
4904 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4910 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4912 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4916 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4918 struct btrfs_root *reloc_root;
4919 struct btrfs_trans_handle *trans;
4920 struct btrfs_key location;
4924 mutex_lock(&root->fs_info->tree_reloc_mutex);
4925 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4927 found = !list_empty(&root->fs_info->dead_reloc_roots);
4928 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4931 trans = btrfs_start_transaction(root, 1);
4933 ret = btrfs_commit_transaction(trans, root);
4937 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4938 location.offset = (u64)-1;
4939 location.type = BTRFS_ROOT_ITEM_KEY;
4941 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4942 BUG_ON(!reloc_root);
4943 btrfs_orphan_cleanup(reloc_root);
4947 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
4948 struct btrfs_root *root)
4950 struct btrfs_root *reloc_root;
4951 struct extent_buffer *eb;
4952 struct btrfs_root_item *root_item;
4953 struct btrfs_key root_key;
4956 BUG_ON(!root->ref_cows);
4957 if (root->reloc_root)
4960 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4963 ret = btrfs_copy_root(trans, root, root->commit_root,
4964 &eb, BTRFS_TREE_RELOC_OBJECTID);
4967 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4968 root_key.offset = root->root_key.objectid;
4969 root_key.type = BTRFS_ROOT_ITEM_KEY;
4971 memcpy(root_item, &root->root_item, sizeof(root_item));
4972 btrfs_set_root_refs(root_item, 0);
4973 btrfs_set_root_bytenr(root_item, eb->start);
4974 btrfs_set_root_level(root_item, btrfs_header_level(eb));
4975 btrfs_set_root_generation(root_item, trans->transid);
4977 btrfs_tree_unlock(eb);
4978 free_extent_buffer(eb);
4980 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4981 &root_key, root_item);
4985 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4987 BUG_ON(!reloc_root);
4988 reloc_root->last_trans = trans->transid;
4989 reloc_root->commit_root = NULL;
4990 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4992 root->reloc_root = reloc_root;
4997 * Core function of space balance.
4999 * The idea is using reloc trees to relocate tree blocks in reference
5000 * counted roots. There is one reloc tree for each subvol, and all
5001 * reloc trees share same root key objectid. Reloc trees are snapshots
5002 * of the latest committed roots of subvols (root->commit_root).
5004 * To relocate a tree block referenced by a subvol, there are two steps.
5005 * COW the block through subvol's reloc tree, then update block pointer
5006 * in the subvol to point to the new block. Since all reloc trees share
5007 * same root key objectid, doing special handing for tree blocks owned
5008 * by them is easy. Once a tree block has been COWed in one reloc tree,
5009 * we can use the resulting new block directly when the same block is
5010 * required to COW again through other reloc trees. By this way, relocated
5011 * tree blocks are shared between reloc trees, so they are also shared
5014 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5015 struct btrfs_root *root,
5016 struct btrfs_path *path,
5017 struct btrfs_key *first_key,
5018 struct btrfs_ref_path *ref_path,
5019 struct btrfs_block_group_cache *group,
5020 struct inode *reloc_inode)
5022 struct btrfs_root *reloc_root;
5023 struct extent_buffer *eb = NULL;
5024 struct btrfs_key *keys;
5028 int lowest_level = 0;
5031 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5032 lowest_level = ref_path->owner_objectid;
5034 if (!root->ref_cows) {
5035 path->lowest_level = lowest_level;
5036 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5038 path->lowest_level = 0;
5039 btrfs_release_path(root, path);
5043 mutex_lock(&root->fs_info->tree_reloc_mutex);
5044 ret = init_reloc_tree(trans, root);
5046 reloc_root = root->reloc_root;
5048 shared_level = ref_path->shared_level;
5049 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5051 keys = ref_path->node_keys;
5052 nodes = ref_path->new_nodes;
5053 memset(&keys[shared_level + 1], 0,
5054 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5055 memset(&nodes[shared_level + 1], 0,
5056 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5058 if (nodes[lowest_level] == 0) {
5059 path->lowest_level = lowest_level;
5060 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5063 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5064 eb = path->nodes[level];
5065 if (!eb || eb == reloc_root->node)
5067 nodes[level] = eb->start;
5069 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5071 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5074 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5075 eb = path->nodes[0];
5076 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5077 group, reloc_inode);
5080 btrfs_release_path(reloc_root, path);
5082 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5088 * replace tree blocks in the fs tree with tree blocks in
5091 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5094 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5095 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5098 extent_buffer_get(path->nodes[0]);
5099 eb = path->nodes[0];
5100 btrfs_release_path(reloc_root, path);
5101 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5103 free_extent_buffer(eb);
5106 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5107 path->lowest_level = 0;
5111 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5112 struct btrfs_root *root,
5113 struct btrfs_path *path,
5114 struct btrfs_key *first_key,
5115 struct btrfs_ref_path *ref_path)
5119 ret = relocate_one_path(trans, root, path, first_key,
5120 ref_path, NULL, NULL);
5126 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5127 struct btrfs_root *extent_root,
5128 struct btrfs_path *path,
5129 struct btrfs_key *extent_key)
5133 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5136 ret = btrfs_del_item(trans, extent_root, path);
5138 btrfs_release_path(extent_root, path);
5142 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5143 struct btrfs_ref_path *ref_path)
5145 struct btrfs_key root_key;
5147 root_key.objectid = ref_path->root_objectid;
5148 root_key.type = BTRFS_ROOT_ITEM_KEY;
5149 if (is_cowonly_root(ref_path->root_objectid))
5150 root_key.offset = 0;
5152 root_key.offset = (u64)-1;
5154 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5157 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5158 struct btrfs_path *path,
5159 struct btrfs_key *extent_key,
5160 struct btrfs_block_group_cache *group,
5161 struct inode *reloc_inode, int pass)
5163 struct btrfs_trans_handle *trans;
5164 struct btrfs_root *found_root;
5165 struct btrfs_ref_path *ref_path = NULL;
5166 struct disk_extent *new_extents = NULL;
5171 struct btrfs_key first_key;
5175 trans = btrfs_start_transaction(extent_root, 1);
5178 if (extent_key->objectid == 0) {
5179 ret = del_extent_zero(trans, extent_root, path, extent_key);
5183 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5189 for (loops = 0; ; loops++) {
5191 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5192 extent_key->objectid);
5194 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5201 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5202 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5205 found_root = read_ref_root(extent_root->fs_info, ref_path);
5206 BUG_ON(!found_root);
5208 * for reference counted tree, only process reference paths
5209 * rooted at the latest committed root.
5211 if (found_root->ref_cows &&
5212 ref_path->root_generation != found_root->root_key.offset)
5215 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5218 * copy data extents to new locations
5220 u64 group_start = group->key.objectid;
5221 ret = relocate_data_extent(reloc_inode,
5230 level = ref_path->owner_objectid;
5233 if (prev_block != ref_path->nodes[level]) {
5234 struct extent_buffer *eb;
5235 u64 block_start = ref_path->nodes[level];
5236 u64 block_size = btrfs_level_size(found_root, level);
5238 eb = read_tree_block(found_root, block_start,
5240 btrfs_tree_lock(eb);
5241 BUG_ON(level != btrfs_header_level(eb));
5244 btrfs_item_key_to_cpu(eb, &first_key, 0);
5246 btrfs_node_key_to_cpu(eb, &first_key, 0);
5248 btrfs_tree_unlock(eb);
5249 free_extent_buffer(eb);
5250 prev_block = block_start;
5253 mutex_lock(&extent_root->fs_info->trans_mutex);
5254 btrfs_record_root_in_trans(found_root);
5255 mutex_unlock(&extent_root->fs_info->trans_mutex);
5256 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5258 * try to update data extent references while
5259 * keeping metadata shared between snapshots.
5262 ret = relocate_one_path(trans, found_root,
5263 path, &first_key, ref_path,
5264 group, reloc_inode);
5270 * use fallback method to process the remaining
5274 u64 group_start = group->key.objectid;
5275 new_extents = kmalloc(sizeof(*new_extents),
5278 ret = get_new_locations(reloc_inode,
5286 ret = replace_one_extent(trans, found_root,
5288 &first_key, ref_path,
5289 new_extents, nr_extents);
5291 ret = relocate_tree_block(trans, found_root, path,
5292 &first_key, ref_path);
5299 btrfs_end_transaction(trans, extent_root);
5305 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5308 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5309 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5311 num_devices = root->fs_info->fs_devices->rw_devices;
5312 if (num_devices == 1) {
5313 stripped |= BTRFS_BLOCK_GROUP_DUP;
5314 stripped = flags & ~stripped;
5316 /* turn raid0 into single device chunks */
5317 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5320 /* turn mirroring into duplication */
5321 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5322 BTRFS_BLOCK_GROUP_RAID10))
5323 return stripped | BTRFS_BLOCK_GROUP_DUP;
5326 /* they already had raid on here, just return */
5327 if (flags & stripped)
5330 stripped |= BTRFS_BLOCK_GROUP_DUP;
5331 stripped = flags & ~stripped;
5333 /* switch duplicated blocks with raid1 */
5334 if (flags & BTRFS_BLOCK_GROUP_DUP)
5335 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5337 /* turn single device chunks into raid0 */
5338 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5343 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5344 struct btrfs_block_group_cache *shrink_block_group,
5347 struct btrfs_trans_handle *trans;
5348 u64 new_alloc_flags;
5351 spin_lock(&shrink_block_group->lock);
5352 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5353 spin_unlock(&shrink_block_group->lock);
5355 trans = btrfs_start_transaction(root, 1);
5356 spin_lock(&shrink_block_group->lock);
5358 new_alloc_flags = update_block_group_flags(root,
5359 shrink_block_group->flags);
5360 if (new_alloc_flags != shrink_block_group->flags) {
5362 btrfs_block_group_used(&shrink_block_group->item);
5364 calc = shrink_block_group->key.offset;
5366 spin_unlock(&shrink_block_group->lock);
5368 do_chunk_alloc(trans, root->fs_info->extent_root,
5369 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5371 btrfs_end_transaction(trans, root);
5373 spin_unlock(&shrink_block_group->lock);
5377 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5378 struct btrfs_root *root,
5379 u64 objectid, u64 size)
5381 struct btrfs_path *path;
5382 struct btrfs_inode_item *item;
5383 struct extent_buffer *leaf;
5386 path = btrfs_alloc_path();
5390 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5394 leaf = path->nodes[0];
5395 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5396 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5397 btrfs_set_inode_generation(leaf, item, 1);
5398 btrfs_set_inode_size(leaf, item, size);
5399 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5400 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5401 btrfs_mark_buffer_dirty(leaf);
5402 btrfs_release_path(root, path);
5404 btrfs_free_path(path);
5408 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5409 struct btrfs_block_group_cache *group)
5411 struct inode *inode = NULL;
5412 struct btrfs_trans_handle *trans;
5413 struct btrfs_root *root;
5414 struct btrfs_key root_key;
5415 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5418 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5419 root_key.type = BTRFS_ROOT_ITEM_KEY;
5420 root_key.offset = (u64)-1;
5421 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5423 return ERR_CAST(root);
5425 trans = btrfs_start_transaction(root, 1);
5428 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5432 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5435 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5436 group->key.offset, 0, group->key.offset,
5440 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5441 if (inode->i_state & I_NEW) {
5442 BTRFS_I(inode)->root = root;
5443 BTRFS_I(inode)->location.objectid = objectid;
5444 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5445 BTRFS_I(inode)->location.offset = 0;
5446 btrfs_read_locked_inode(inode);
5447 unlock_new_inode(inode);
5448 BUG_ON(is_bad_inode(inode));
5452 BTRFS_I(inode)->index_cnt = group->key.objectid;
5454 err = btrfs_orphan_add(trans, inode);
5456 btrfs_end_transaction(trans, root);
5460 inode = ERR_PTR(err);
5465 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5468 struct btrfs_ordered_sum *sums;
5469 struct btrfs_sector_sum *sector_sum;
5470 struct btrfs_ordered_extent *ordered;
5471 struct btrfs_root *root = BTRFS_I(inode)->root;
5472 struct list_head list;
5477 INIT_LIST_HEAD(&list);
5479 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5480 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5482 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5483 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5484 disk_bytenr + len - 1, &list);
5486 while (!list_empty(&list)) {
5487 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5488 list_del_init(&sums->list);
5490 sector_sum = sums->sums;
5491 sums->bytenr = ordered->start;
5494 while (offset < sums->len) {
5495 sector_sum->bytenr += ordered->start - disk_bytenr;
5497 offset += root->sectorsize;
5500 btrfs_add_ordered_sum(inode, ordered, sums);
5502 btrfs_put_ordered_extent(ordered);
5506 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5508 struct btrfs_trans_handle *trans;
5509 struct btrfs_path *path;
5510 struct btrfs_fs_info *info = root->fs_info;
5511 struct extent_buffer *leaf;
5512 struct inode *reloc_inode;
5513 struct btrfs_block_group_cache *block_group;
5514 struct btrfs_key key;
5523 root = root->fs_info->extent_root;
5525 block_group = btrfs_lookup_block_group(info, group_start);
5526 BUG_ON(!block_group);
5528 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5529 (unsigned long long)block_group->key.objectid,
5530 (unsigned long long)block_group->flags);
5532 path = btrfs_alloc_path();
5535 reloc_inode = create_reloc_inode(info, block_group);
5536 BUG_ON(IS_ERR(reloc_inode));
5538 __alloc_chunk_for_shrink(root, block_group, 1);
5539 set_block_group_readonly(block_group);
5541 btrfs_start_delalloc_inodes(info->tree_root);
5542 btrfs_wait_ordered_extents(info->tree_root, 0);
5547 key.objectid = block_group->key.objectid;
5550 cur_byte = key.objectid;
5552 trans = btrfs_start_transaction(info->tree_root, 1);
5553 btrfs_commit_transaction(trans, info->tree_root);
5555 mutex_lock(&root->fs_info->cleaner_mutex);
5556 btrfs_clean_old_snapshots(info->tree_root);
5557 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5558 mutex_unlock(&root->fs_info->cleaner_mutex);
5560 trans = btrfs_start_transaction(info->tree_root, 1);
5561 btrfs_commit_transaction(trans, info->tree_root);
5564 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5568 leaf = path->nodes[0];
5569 nritems = btrfs_header_nritems(leaf);
5570 if (path->slots[0] >= nritems) {
5571 ret = btrfs_next_leaf(root, path);
5578 leaf = path->nodes[0];
5579 nritems = btrfs_header_nritems(leaf);
5582 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5584 if (key.objectid >= block_group->key.objectid +
5585 block_group->key.offset)
5588 if (progress && need_resched()) {
5589 btrfs_release_path(root, path);
5596 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5597 key.objectid + key.offset <= cur_byte) {
5603 cur_byte = key.objectid + key.offset;
5604 btrfs_release_path(root, path);
5606 __alloc_chunk_for_shrink(root, block_group, 0);
5607 ret = relocate_one_extent(root, path, &key, block_group,
5613 key.objectid = cur_byte;
5618 btrfs_release_path(root, path);
5621 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5622 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5625 if (total_found > 0) {
5626 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5627 (unsigned long long)total_found, pass);
5629 if (total_found == skipped && pass > 2) {
5631 reloc_inode = create_reloc_inode(info, block_group);
5637 /* delete reloc_inode */
5640 /* unpin extents in this range */
5641 trans = btrfs_start_transaction(info->tree_root, 1);
5642 btrfs_commit_transaction(trans, info->tree_root);
5644 spin_lock(&block_group->lock);
5645 WARN_ON(block_group->pinned > 0);
5646 WARN_ON(block_group->reserved > 0);
5647 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5648 spin_unlock(&block_group->lock);
5649 put_block_group(block_group);
5652 btrfs_free_path(path);
5656 static int find_first_block_group(struct btrfs_root *root,
5657 struct btrfs_path *path, struct btrfs_key *key)
5660 struct btrfs_key found_key;
5661 struct extent_buffer *leaf;
5664 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5669 slot = path->slots[0];
5670 leaf = path->nodes[0];
5671 if (slot >= btrfs_header_nritems(leaf)) {
5672 ret = btrfs_next_leaf(root, path);
5679 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5681 if (found_key.objectid >= key->objectid &&
5682 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5693 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5695 struct btrfs_block_group_cache *block_group;
5696 struct btrfs_space_info *space_info;
5699 spin_lock(&info->block_group_cache_lock);
5700 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5701 block_group = rb_entry(n, struct btrfs_block_group_cache,
5703 rb_erase(&block_group->cache_node,
5704 &info->block_group_cache_tree);
5705 spin_unlock(&info->block_group_cache_lock);
5707 btrfs_remove_free_space_cache(block_group);
5708 down_write(&block_group->space_info->groups_sem);
5709 list_del(&block_group->list);
5710 up_write(&block_group->space_info->groups_sem);
5712 WARN_ON(atomic_read(&block_group->count) != 1);
5715 spin_lock(&info->block_group_cache_lock);
5717 spin_unlock(&info->block_group_cache_lock);
5719 /* now that all the block groups are freed, go through and
5720 * free all the space_info structs. This is only called during
5721 * the final stages of unmount, and so we know nobody is
5722 * using them. We call synchronize_rcu() once before we start,
5723 * just to be on the safe side.
5727 while(!list_empty(&info->space_info)) {
5728 space_info = list_entry(info->space_info.next,
5729 struct btrfs_space_info,
5732 list_del(&space_info->list);
5738 int btrfs_read_block_groups(struct btrfs_root *root)
5740 struct btrfs_path *path;
5742 struct btrfs_block_group_cache *cache;
5743 struct btrfs_fs_info *info = root->fs_info;
5744 struct btrfs_space_info *space_info;
5745 struct btrfs_key key;
5746 struct btrfs_key found_key;
5747 struct extent_buffer *leaf;
5749 root = info->extent_root;
5752 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5753 path = btrfs_alloc_path();
5758 ret = find_first_block_group(root, path, &key);
5766 leaf = path->nodes[0];
5767 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5768 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5774 atomic_set(&cache->count, 1);
5775 spin_lock_init(&cache->lock);
5776 mutex_init(&cache->alloc_mutex);
5777 mutex_init(&cache->cache_mutex);
5778 INIT_LIST_HEAD(&cache->list);
5779 read_extent_buffer(leaf, &cache->item,
5780 btrfs_item_ptr_offset(leaf, path->slots[0]),
5781 sizeof(cache->item));
5782 memcpy(&cache->key, &found_key, sizeof(found_key));
5784 key.objectid = found_key.objectid + found_key.offset;
5785 btrfs_release_path(root, path);
5786 cache->flags = btrfs_block_group_flags(&cache->item);
5788 ret = update_space_info(info, cache->flags, found_key.offset,
5789 btrfs_block_group_used(&cache->item),
5792 cache->space_info = space_info;
5793 down_write(&space_info->groups_sem);
5794 list_add_tail(&cache->list, &space_info->block_groups);
5795 up_write(&space_info->groups_sem);
5797 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5800 set_avail_alloc_bits(root->fs_info, cache->flags);
5801 if (btrfs_chunk_readonly(root, cache->key.objectid))
5802 set_block_group_readonly(cache);
5806 btrfs_free_path(path);
5810 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5811 struct btrfs_root *root, u64 bytes_used,
5812 u64 type, u64 chunk_objectid, u64 chunk_offset,
5816 struct btrfs_root *extent_root;
5817 struct btrfs_block_group_cache *cache;
5819 extent_root = root->fs_info->extent_root;
5821 root->fs_info->last_trans_new_blockgroup = trans->transid;
5823 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5827 cache->key.objectid = chunk_offset;
5828 cache->key.offset = size;
5829 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5830 atomic_set(&cache->count, 1);
5831 spin_lock_init(&cache->lock);
5832 mutex_init(&cache->alloc_mutex);
5833 mutex_init(&cache->cache_mutex);
5834 INIT_LIST_HEAD(&cache->list);
5836 btrfs_set_block_group_used(&cache->item, bytes_used);
5837 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5838 cache->flags = type;
5839 btrfs_set_block_group_flags(&cache->item, type);
5841 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5842 &cache->space_info);
5844 down_write(&cache->space_info->groups_sem);
5845 list_add_tail(&cache->list, &cache->space_info->block_groups);
5846 up_write(&cache->space_info->groups_sem);
5848 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5851 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5852 sizeof(cache->item));
5855 set_avail_alloc_bits(extent_root->fs_info, type);
5860 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5861 struct btrfs_root *root, u64 group_start)
5863 struct btrfs_path *path;
5864 struct btrfs_block_group_cache *block_group;
5865 struct btrfs_key key;
5868 root = root->fs_info->extent_root;
5870 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5871 BUG_ON(!block_group);
5872 BUG_ON(!block_group->ro);
5874 memcpy(&key, &block_group->key, sizeof(key));
5876 path = btrfs_alloc_path();
5879 spin_lock(&root->fs_info->block_group_cache_lock);
5880 rb_erase(&block_group->cache_node,
5881 &root->fs_info->block_group_cache_tree);
5882 spin_unlock(&root->fs_info->block_group_cache_lock);
5883 btrfs_remove_free_space_cache(block_group);
5884 down_write(&block_group->space_info->groups_sem);
5885 list_del(&block_group->list);
5886 up_write(&block_group->space_info->groups_sem);
5888 spin_lock(&block_group->space_info->lock);
5889 block_group->space_info->total_bytes -= block_group->key.offset;
5890 block_group->space_info->bytes_readonly -= block_group->key.offset;
5891 spin_unlock(&block_group->space_info->lock);
5892 block_group->space_info->full = 0;
5894 put_block_group(block_group);
5895 put_block_group(block_group);
5897 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5903 ret = btrfs_del_item(trans, root, path);
5905 btrfs_free_path(path);